AI Tools Directory

Google DeepMind's research initiative focused on building world models that can generate interactive, playable 3D environments, with near-term applications in gaming.

An AI system that builds controllable 3D digital twins from real-world data like point clouds or depth video, modeling not just object appearance but also how objects move, bend, and deform.

NVIDIA's 3D reconstruction model that takes multiple images of a scene and reconstructs the entire scene in 3D as Gaussian splats, designed to be small and efficient compared to traditional 3D reconstruction transformers.

A 3D reconstruction model that can turn a set of images into simulation-ready 3D scenes, enabling accurate articulation and animation of reconstructed objects.

Apple's 3D model generator that reconstructs view-dependent 3D representations from input images, capturing how objects visually behave from different angles including reflections and lighting changes.

An open-source image-to-3D model generator that creates highly accurate 3D meshes from single images using pixel-aligned reconstruction, producing detailed geometry and realistic textures superior to competitors.

A 3D model generation tool that converts images into 3D models, developed as a competitor in the image-to-3D generation space.

A 3D asset generation model that creates 3D models from images, serving as one of the competitors in the image-to-3D generation space.

A Google DeepMind model announced in August that can generate dynamic, playable worlds from text prompts or images, retaining consistency for a few minutes at 720p resolution.

A compute marketplace platform that tracks and provides pricing data for GPU compute resources, including spot prices for Nvidia H100 and other accelerator hardware.

A standardized JSON schema and datastore for reporting, storing, and comparing AI model evaluation results across benchmarks, harnesses, and leaderboards. It aggregates around 229,000 evaluation results across more than 22,000 models and 2,200 benchmarks from 31 different reporting formats.

A platform for browsing and verifying AI model evaluation results, providing source badges and cross-linking to full structured evaluation records from the Every Eval Ever datastore.

An AI leaderboard platform originating from UC Berkeley that crowdsources over 10 million user evaluations to rank AI models across text, coding, vision, image generation, and agent tasks. It also offers Arena Pro, a commercial analytics service for model labs and enterprises.

A multi-AI assistant platform that provides access to multiple leading AI models including GPT, Claude, Gemini, and Sonar under one roof, along with AI image generation, video creation tools, document analysis, OCR text extraction, and real-time web search. It works across iOS, Android, and the web through a single account.

The first open, community-driven benchmark by Treble Technologies and Hugging Face for evaluating automatic speech recognition models under realistic far-field acoustic conditions, measuring performance across varying room acoustics, noise levels, and microphone distances.

Hugging Face's reference implementation of the Agentic Resource Discovery (ARD) specification, providing search access to thousands of Skills, ML applications, and MCP Servers on the Hugging Face Hub and across other ARD discovery services.

A company that tracks and benchmarks AI model performance, providing independent evaluations and comparisons of frontier AI models from various providers.

A benchmarking index by Artificial Analysis that evaluates and ranks AI coding models based on their performance across coding tasks.

A benchmarking platform (formerly known as LMSYS Chatbot Arena) that ranks and compares large language models based on human preference evaluations, maintaining a leaderboard chart of model performance.

An open benchmark by IBM Research and Hugging Face for comparing full AI agent systems across multiple tasks, reporting both quality and cost metrics to evaluate generality and deployment worthiness.

The Massive Text Embedding Benchmark is a comprehensive evaluation framework for assessing text embedding models across multiple tasks including retrieval, classification, and clustering in both English and multilingual settings.

A monthly index published by fintech company Ramp that tracks AI adoption among businesses by analyzing expense data from over 50,000 companies using Ramp's platform.

A relaunched news aggregator platform that ingests content from X (formerly Twitter) in real time, performing sentiment analysis, clustering, and signal detection to surface and rank the most important AI news stories.

A built-in profiling tool in PyTorch that allows developers to analyze and optimize the performance of their deep learning models by generating detailed traces and tables of operations, kernel launches, and GPU utilization.

Nvidia's parallel computing platform and programming model that enables developers to use Nvidia GPUs for general-purpose computing, widely adopted as the default engine for AI research and deep learning workloads.

A built-in analytics dashboard feature within Anthropic's Claude that lets users track and visualize their AI usage patterns, topics discussed, and task types, while offering mindful usage tools like quiet hours and break nudges.

A multimodal AI model by Meta designed for agentic coding, capable of multistep reasoning, managing digital workflows, handling large code migrations, fixing bugs, and deploying new features in enterprise systems. It is priced at $1.25 per million input tokens and $4.25 per million output tokens.

An AI workplace agent by OpenAI designed for enterprise teams, capable of handling multistep office tasks such as drafting documents, spreadsheets, and presentations across desktop, web, and mobile.

A Google Chrome browser extension by OpenAI that gives ChatGPT access to the context of web pages being viewed, allowing users to ask questions, summarize content, and initiate tasks directly from the browser.

An AI agent built by Lyzr that autonomously managed the startup's $100 million fundraise, fielding questions from over 130 investors, drafting investment memos, and tracking investor engagement with presentation slides.

An interactive visual map tool that allows users to explore and navigate NVIDIA's Nemotron v3 post-training data collection, with color overlays and filters to reorganize prompts by dataset, pipeline stage, domain, or tool use.

Hugging Face's reference modeling library for machine learning that supports 450+ architectures through consistent APIs, now integrated as a modeling backend in vLLM to enable native-speed inference without custom model porting.

An AI assistant startup that raised $40 million in seed funding from DST and Lux Capital, focusing on visual responses to make AI assistants more interactive and engaging.

A startup providing a full-stack platform for AI agent development, including compute access, reinforcement learning frameworks, and evaluation tools. It enables enterprises to train their own agentic AI systems without relying on frontier AI labs.

Samsung's AI-powered photo editing feature built into Galaxy smartphones that allows users to edit images by typing or tapping, enabling detail restoration, object removal, and application of filters or creative styles.

An integrated development environment by AWS for machine learning that provides tools for building, training, fine-tuning, and deploying ML models with pre-configured environments and GPU management.

A machine learning hub within Amazon SageMaker that provides pre-trained foundation models, built-in algorithms, and solution templates to accelerate ML development and deployment.

An AWS service for deploying machine learning models to production endpoints, supporting real-time and batch inference with managed infrastructure and scaling.

A zero-egress storage service by Hugging Face that allows users to store and access models, datasets, and checkpoints from any cloud without transfer fees, using content-defined chunking for efficient deduplication.

A managed GPU platform-as-a-service within Microsoft Foundry for deploying open-source and custom AI models, handling container updates, runtime upgrades, and security patches automatically.

A Microsoft service within the Foundry platform that provides multi-agent orchestration with built-in memory, knowledge grounding through Foundry IQ, and a catalog of connectable tools via agentic protocols.

An always-on AI assistant by Anthropic that integrates into Slack, acting as an AI teammate that can participate in workspace conversations and assist with tasks.

A consumer security app for iOS and Android that uses AI to screen texts, voicemails, and incoming phone calls for scams, featuring live-call monitoring that can detect behavioral tells of fraud in real time.

A free, anonymous web tool by Savi that allows users to upload suspicious texts, photos, or emails to determine if they are likely scams, using AI-based scam detection models.

A high-performance LLM inference server by French startup ZML that enables running open-source large language models across a variety of AI chips including Nvidia, AMD, Google TPU, Apple Metal, and Intel Arc, aiming to break vendor lock-in and maximize chip performance.

A reward model for robotics integrated into LeRobot's reward models API, designed to evaluate when a robot has successfully completed a task.

A reward model for robotics integrated into LeRobot's reward models API, used to assess robot task success.

A cloud training service by Hugging Face that enables users to run model training jobs in the cloud, integrated with frameworks like LeRobot for distributed training workflows.

A dataset format designed for distributed training that supports efficient streaming, mixing, and shuffling of data, including training directly from object storage like S3 or GCS.

Vercel's AI gateway service that routes AI model requests, allowing developers to plug in different AI providers like OpenAI, Anthropic, or Gemini, with over 1 trillion tokens flowing through it.

A framework by Vercel for building and managing AI agents, allowing users to define an agent's instructions and skills in natural language for both customer-facing and internal productivity use cases.

A secure execution environment by Vercel that constrains AI agents within controlled boundaries, applying policies on what data agents can access and what data can leave the sandbox to prevent data leakage.

A Microsoft Excel add-in that integrates ChatGPT directly into Excel, allowing users to create spreadsheets from scratch, fill in data from uploaded files, automate edits, and perform complex spreadsheet tasks using natural language prompts with adjustable reasoning levels.

A Hugging Face project that standardizes how custom GPU kernels are packaged, distributed, and consumed on the Hugging Face Hub. It provides a dedicated repository type for kernels with security features including trusted publishers, code signing, and reproducible builds.

An open-source code signing tool from the Sigstore project that uses ephemeral private keys to sign software artifacts, providing supply chain security and provenance verification.

AMD's line of data center GPU accelerators designed for AI training and inference workloads, serving as an alternative to Nvidia's GPU offerings for high-performance computing.

A Google feature accessible via Google Search, YouTube, and Google Discover that lets users learn about ads, including a new disclosure indicating whether an ad was created or edited with AI.

A completely AI-free search product from DuckDuckGo designed for users who want a privacy-focused search experience without any AI features.

An AI-powered browser that lets users ask questions about any web page, uncover relevant insights, receive personalized recommendations based on browsing activity, and includes an integrated Notes app with AI summarization capabilities.

A privacy-first web browser with built-in ad and tracker blocking capabilities, a VPN service, and a gamified browsing model that rewards users with Basic Attention Token (BAT) cryptocurrency for opting in to view ads.

An agentic business workspace designed for solopreneurs and small business owners that automates tasks like setting up Shopify stores, managing email communications, scheduling reports, and conducting browser-based research.

Google's subscription tier that provides access to advanced AI features across Google's consumer apps, including generative AI tools in Google Photos.

A marketplace built by General Intuition that connects gamers to data labeling and teleoperations work, aimed at addressing AI-driven job displacement.

An AI startup that helps enterprises develop and deploy AI agents for software development workflows, valued at $1.5 billion.

A public tool by OpenAI that allows users to upload images and check whether they contain SynthID watermarks, helping verify if content was AI-generated.

NVIDIA's open-source library for optimizing large language model inference on NVIDIA GPUs, providing state-of-the-art performance through kernel fusion, quantization, and parallelism techniques.

NVIDIA Inference Microservices (NIM) is a set of optimized containerized microservices by NVIDIA for deploying AI models with high performance across cloud, data center, and workstation environments.

Text Embeddings Inference (TEI) is a Hugging Face toolkit for deploying and serving text embedding models with high performance, supporting popular embedding and reranking models.

A cloud-based AI inference platform by SambaNova Systems launched in September 2023, providing access to SambaNova's custom AI chip infrastructure for running large language models.

Cloudflare's suite of AI products and tools that help users launch their own AI systems, combat AI bots, and manage content access through features like Pay Per Crawl and Pay Per Use for AI crawlers.

Samsung's AI-powered camera feature that uses machine learning to reduce noise and balance exposure in low-light photography and videography on Galaxy smartphones.

An open-source speculative decoder system developed by DeepSeek that can speed up AI inference by up to 85% during testing on small models.

A Chinese AI company that develops and hosts large language models including the GLM series, offering both a web-based chat interface and an API for developers to integrate their models into applications.

A popular open-source tool for building LLM-powered applications, providing a visual interface for creating and deploying language model workflows.

A cloud platform providing AI infrastructure and GPU compute services for training and deploying machine learning models.

An open benchmark by IBM Research for evaluating AI coding agents on cross-framework migration tasks in Enterprise Java, testing whether migrated applications actually build, deploy, and preserve behavior across Spring, Jakarta EE, and Quarkus ecosystems.

A Singapore-based startup offering an agentic AI keyboard for Android that embeds AI capabilities directly into the smartphone keyboard, enabling users to take actions like translations, recommendations, and information lookups without switching apps. It is powered by Google's Gemini models and features customizable 'Skills' shortcuts.

A hosted Model Context Protocol (MCP) server by X (formerly Twitter) that allows AI assistants and MCP-compatible applications to connect directly to the X API using a user's own account permissions, enabling search, reading posts, user lookups, and conversation analysis without developers needing to build their own integration infrastructure.

A Prague-based AI lab founded by former DeepMind researchers that applies reinforcement learning techniques to stock and crypto trading. Its algorithms trade billions in daily volume across the S&P 500 and Nasdaq in partnership with quant firm Tower Research Capital.

An AI chip startup founded in 2022 that designs specialized inference chips and frontier inference clusters — bundled systems including custom chips, racks, and software — built to run frontier AI models faster and more cheaply than general-purpose GPUs.

A comprehensive cybersecurity suite by Bitdefender that covers up to 15 or 25 devices under a single subscription, providing automated threat detection across Windows, macOS, iOS, and Android systems with features including parental controls, password manager, anti-tracker, and the sandboxed Safepay browser.

Bitdefender's centralized management console that allows users to configure security settings, parental controls, and device management across all connected devices from a single dashboard.

A fully managed machine learning service by Amazon Web Services that enables developers and data scientists to build, train, and deploy ML models at scale, including data annotation workflows.

Omen AI develops tiny spectrometers with AI-powered signal processing that monitor fluid health in real time within data centers and heavy machinery, detecting bacterial growth and equipment wear before they cause costly downtime.

Alibaba's internal AI coding tool, positioned as an alternative to external programming assistants like Claude Code for the company's employees.

An open-source code agent developed by Mistral AI for AI-assisted software development tasks.

Google's smart TV operating system integrated with Gemini AI capabilities, providing an intelligent interface for content discovery and TV interaction on supported television sets.

Tesla's advanced driver-assistance system that uses AI and computer vision to enable semi-autonomous driving features including Autopark and navigation assistance.

A marketplace built by crypto exchange OKX where AI agents can discover, hire, and pay one another autonomously using blockchain-based payments and stablecoins, while building portable on-chain reputations.

Anthropic's native integration of Claude into Slack that allows users to tag Claude from any channel or conversation, initiating background work through a full instance of Claude Code within the team's communication flow.

An autonomous software factory built on the Hermes Agent platform that scopes, builds, and deploys real software projects without human intervention, generating deployed URLs from simple idea descriptions.

A Meta app that allows users to generate small, interactive apps and games using AI prompts, featuring a scrollable feed for discovering and playing with creations made by others.

A livestreaming command center on X (formerly Twitter) that allows creators to go live, create test streams, schedule broadcasts, use RTMP for external software, and monitor live audience data. It replaces the previous Media Studio Producer.

An AI-enabled feature on Meta's smart glasses that amplifies the voices of people the wearer is facing, making communication easier in noisy environments. It runs entirely on-device on the glasses themselves.

A Mac utility application by MacPaw that helps users clean system junk, find duplicate files, monitor system health, manage storage, and perform malware scans through features like Smart Care, Space Lens, and Cloud Cleanup.

A chat interface by Zhipu AI for accessing GLM family models, offering features like DeepThink modes (High and Max), advanced multi-round search, in-depth research analysis, agent capabilities for complex tasks, and code generation for landing pages and applications.

A Google-provided bot control mechanism that allows website owners to opt out of having their content used for Google's AI training and AI products like Gemini Apps and Vertex API, without affecting Google Search inclusion.

An AI search platform that partners with Cloudflare to enable publishers to get paid when their content appears in AI-generated search results.

An open-source data integration platform that helps organizations move and consolidate data from various sources into data warehouses, lakes, and other destinations.

An AI-powered automation platform that allows users to build AI agents and workflows for various business tasks, integrating with multiple AI models and services.

A network automation platform that provides software for configuring and operating GPU cluster networking at data centers, helping AI neoclouds reduce time to go live through automated setup, hardware-accelerated network abstraction, and multi-tenancy support.

An AI-powered image and video enhancement tool maker offering models like Astra for video upscaling and Wonder for image retouching, known for optimizing large AI models to run on consumer-grade GPUs. Acquired by Adobe in 2026.

A startup founded by former Meta AI researchers that builds simulated digital environments ('digital world models') to stress-test and evaluate AI agents using reinforcement learning, serving major AI labs and enterprises.

A mobile document scanning app for iPhone and iPad that uses AI-powered border detection and adjustment to scan, edit, and manage documents. It supports exporting to multiple formats (PDF, JPG, DOC, XLS, PPT, TXT), includes OCR capabilities, PDF editing, electronic signatures, and document organization features.

An all-in-one PDF editor for Windows and Mac that allows users to edit, convert, merge, split, compress, and sign PDFs. It includes OCR technology to convert scanned documents and images into editable text, along with document comparison, redaction tools, and password protection.

An open-source deep learning library created by François Chollet that provides a high-level API for building and training neural networks, widely used for rapid prototyping and research.

A routing layer for the OpenClaw platform that connects user requests to AI agents and the tools and skills those agents use to complete tasks.

An enterprise automation platform that uses AI and robotic process automation (RPA) to help organizations automate business processes and workflows.

A training infrastructure system used in the development of GLM models that enables many long-running coding agents to practice in parallel without system breakdowns, supporting reinforcement learning at scale.

An open library within the NVIDIA NeMo framework that accelerates fine-tuning of large language models, particularly Mixture-of-Experts (MoE) architectures, by adding Expert Parallelism, DeepEP fused all-to-all dispatch, and TransformerEngine kernels on top of HuggingFace Transformers v5, delivering 3.4-3.7x higher training throughput with 29-32% less GPU memory.

A library by NVIDIA that provides optimized kernels for transformer model training, including support for mixed-precision computation and fused operations to accelerate deep learning workloads on NVIDIA GPUs.

A communication library for Mixture-of-Experts models that provides fused all-to-all dispatch, overlapping communication with expert computation to improve distributed training efficiency.

An acoustic simulation engine by Treble Technologies that combines wave-based solvers at low to mid frequencies with geometrical-acoustics modeling at higher frequencies to generate realistic room impulse responses for audio evaluation.

An enterprise AI services startup founded by former Infosys CEO Vishal Sikka that uses agentic code generation, reusable AI skills, and domain expertise to help enterprises build, modify, and operate software through AI-driven development and automation.

A standalone AI companion app by Meta designed to help Facebook creators grow their audiences by providing personalized recommendations, AI-powered comment management, and performance insights based on their content and engagement data.

An AI assistant built into Facebook's Creator Studio that provides creators with personalized recommendations, answers performance questions conversationally, and helps analyze audience engagement and content strategy.

An AI-powered software suite built into Acer laptops that provides quick access to AI tools for tweaking settings, enhancing video calls, and adjusting system performance through a single shortcut.

TensorWave is a cloud computing platform that provides AI compute infrastructure built on AMD GPUs, offering scalable GPU clusters for AI training and inference workloads.

A blockchain security platform that offers AI-powered services to assess the security of crypto wallets and tokens before executing transactions.

A crypto market data service that provides live market data on a pay-per-query basis, designed for integration with AI agents and autonomous systems.

A dispute-resolution infrastructure platform built by GenLayer Labs that helps AI agents resolve contractual disagreements, functioning as a digital court system for autonomous transactions.

An AI tool built using Anthropic's Claude, created specifically for California state workers to assist with common business needs and government workflows.

An open-source agent harness from IBM Research that handles planning, execution loops, tool calls, and state management for building agentic AI applications. It supports interchangeable tools (OpenAPI, MCP, LangChain), long-horizon planning with self-correction, declarative guardrails, multi-agent delegation, and one-env-var provider switching across OpenAI, Anthropic, watsonx, Ollama, and more.

A Stockholm-based video-first hiring platform that uses AI interview agents to conduct personalized candidate interviews and generate short-form video profiles, creating a browsable talent pool for employers.

An AI-powered marketing platform that assigns a dedicated AI agent to each individual customer, enabling brands to personalize messaging based on individual behavior rather than traditional audience segments and campaign rules.

A customer engagement platform that serves over 1,350 consumer brands across 75 countries, providing marketing automation and personalized messaging capabilities, now integrating AI agent technology through its acquisition of Aampe.

The latest generation of PaddleOCR's universal OCR model family, supporting text detection and recognition across 50 languages with model sizes ranging from 1.5M to 34.5M parameters. It achieves 86.2% detection Hmean and 83.2% recognition accuracy, designed for real-world deployment across documents, screenshots, multilingual images, and scene text.

A Samsung Galaxy AI content creation tool that works alongside Photo Assist and Bixby to enable on-device editing and creative content generation on Galaxy S26 devices.

A crowdsourcing marketplace by Amazon where human workers perform microtasks such as data annotation, sentiment labeling, and CAPTCHA completion — tasks historically used to generate training data for AI and machine learning models.

An AI-powered cybersecurity tool developed by Chinese firm 360 that is designed to automatically discover software vulnerabilities, positioned as a competitor to Anthropic's Mythos.

An AI-powered cybersecurity tool developed by Chinese firm 360 that automates cyber defense and incident response operations.

Smart glasses developed through a partnership between Ray-Ban and Meta, featuring a built-in camera, speakers, and Meta AI assistant capabilities in a classic Wayfarer frame design.

A website created by Thomas Dimson and Joey Flynn that queries multiple AI models to measure how well they can recall a given person without web search, assigning a strength score to indicate how prominently someone is encoded in AI model weights.

A creative platform by Google that allows users to build custom AI-powered creative tools using plain English instructions. It includes pre-built tools for image generation, video editing, prompt organization, and experimental features like 4D depth visualization, with the ability to remix and create new tools through natural language.

An AI assistant developed by Reliance Jio that can join phone calls to transcribe conversations, generate summaries, and perform tasks such as booking cabs, ordering food, and making reservations. It is activated by saying 'Hey Jio' and is designed for Jio's 500+ million users.

An AI-enhanced version of Reliance Jio's mobile app that can perform tasks on behalf of users through natural-language requests, such as activating eSIMs and selecting roaming plans.

A home display device by Reliance that uses AI agents to proactively surface information and recommendations such as weather alerts, schedules, and household reminders, functioning as an ambient AI assistant for the home.

Reliance Industries' AI platform aimed at developing AI infrastructure and services for consumers, businesses, and governments, including applications supporting 22 Indian languages.

An enterprise AI startup founded by Vishal Sikka that focuses on AI applications and analytics tools designed to help businesses use artificial intelligence in decision-making.

An Amazon smart display with Alexa voice assistant built in, featuring an 11-inch screen for video calls, smart home control, and media consumption.

Amazon's most powerful streaming stick featuring Alexa voice assistant integration, Wi-Fi 6E support, and ambient smart display capabilities.

A Hugging Face coding agent framework that enables AI agents to write, run, and debug code autonomously to solve tasks using various tools and libraries.

A wellness app created by Karamo Brown that offers personalized fitness plans, nutrition guidance, meditation content, and an AI digital clone of Brown that users can interact with for real-time advice.

An AI startup that creates digital clones of real people by training on their interviews, podcast episodes, and other content, enabling lifelike conversational replicas used in apps and platforms.

An AI-powered mind mapping and knowledge organization tool that automatically transforms content from PDFs, videos, websites, and other sources into structured mind maps, summaries, and visual diagrams.

An Apple Intelligence-powered real-time translation feature that provides seamless, hands-free communication in over 10 languages through compatible AirPods and iPhones running iOS 26.

A document processing tool integrated into CUGA that powers retrieval-augmented generation (RAG) capabilities for agentic applications.

A cloud-based code execution sandbox service that provides secure environments for running AI-generated code, supported as a sandbox option in the CUGA agent harness.

IBM's enterprise AI and data platform that provides foundation models, generative AI, and machine learning capabilities, supported as a model provider in CUGA.

Salesforce's enterprise marketing automation platform that enables brands to manage customer journeys, email campaigns, and personalized marketing across multiple channels.

An open-source SDK from AWS that exposes robot abstractions, simulation, and the LeRobot stack as composable AgentTools within a single agent loop, enabling orchestration of robot demonstrations, training, and deployment.

An open draft specification developed by contributors from Microsoft, Google, GoDaddy, Hugging Face, and others that defines how AI agents, tools, and skills are cataloged, indexed, and searched across federated registries, enabling runtime capability discovery.

A startup that combines conventional LLMs with formal verification techniques inspired by the LEAN programming language to provide deterministic reliability layers for AI in sensitive domains like law, drug discovery, and tax preparation.

A real-time audio streaming platform originally designed for concert experiences that evolved to power live-event audio with integrated translation capabilities. Founded in 2016, it was acquired by DeepL to enhance live-event translation experiences.

An AI model by Pinterest that helps advertisers automatically select the best-performing ad creative from multiple options each time an ad is shown, optimizing campaign performance globally.

AI-powered algorithm personalization features on Meta's Threads platform that let users influence their feed recommendations by specifying topics they want to see more or less of, either publicly ('Dear Algo') or privately ('Your Algo').

Google's new $99 smart speaker built with Gemini AI integration, capable of understanding complex natural-language commands for smart home control, launching June 25, 2025 in four colors.

A screen-time reduction app created by JAMDAT founder Mitch Lasky (correction: founded by a mobile gaming pioneer) designed to help users reduce phone addiction through product design rather than willpower.

An app designed to limit screen time and social media usage by introducing intentional pauses before opening distracting apps.

A startup building data pipelines, collection tools, and annotation systems for robot training data, providing the physical-world data infrastructure that frontier AI labs and robotics companies need to train embodied AI models.

A low-cost teleoperation system developed at UC Berkeley that allows a human operator to control a robotic arm to generate robot training data.

An open-source, self-hosted AI workspace created by PewDiePie that serves as an interface for local and API-based AI models, offering features like chatting, agents, deep research, document writing, file management, and integrations with email and calendars.

A cross-platform inference engine by Microsoft for running machine learning models in the ONNX format, enabling portable deployment across various hardware and software environments.

A service by Hugging Face that provides access to various AI model inference endpoints, enabling developers to run open-weight models through a managed API infrastructure.

An inference hardware system by NVIDIA designed for high-performance AI model inference, announced at NVIDIA's GTC event and built using technology licensed from Groq's language processing unit IP.

A custom AI accelerator chip developed by Google specifically for machine learning workloads, used to train and run Google's AI models at scale.

An AI-powered hardware and software platform that creates notetaking devices (Plaud Pin, Plaud Pro, Plaud Pin S) for recording and transcribing real-world conversations, generating summaries and action items. Its software subscription business has surpassed $100M in annualized revenue.

An AI startup that builds a data science tool designed to prevent hallucinations and factual errors in LLM outputs by using a deterministic validator system and harness engineering. It aims for 99.99% accuracy and runs on smaller models to reduce token costs.

Walmart's AI shopping assistant embedded in its platform to help shoppers find products, competing with similar AI assistants from other retailers.

A smart speaker and alarm clock combo by Amazon with Alexa voice assistant built in, featuring customizable clock faces and sounds in a compact nightstand-friendly design.

An agentic coding framework developed by Zhipu AI (ZAI) that allows users to run GLM models in a project-based coding environment similar to OpenAI's Codex, supporting Mac, Windows, and Linux.

A cybersecurity startup that provides an identity management platform designed to authenticate, govern, and control both human and AI-agent identities in a unified system, featuring split-key architecture and integrations for coding assistants.

A software package developed by NASA's Jet Propulsion Laboratory that serves as a harness for deploying vision-language models on satellites, enabling onboard AI-powered analysis of Earth observation data in response to natural language queries.

A customer conversation management platform based in Malaysia that helps mid- to large-sized B2C businesses manage customer interactions across multiple messaging channels including WhatsApp, Instagram, TikTok, and more. It uses AI agents to automatically handle high volumes of customer inquiries, qualify leads, and close sales without human intervention.

XPENG's Vision-Language-Action 2.0 autonomous driving system, an AI-powered model with L4 autonomous driving potential that uses physical AI to navigate complex real-world driving scenarios.

An AI infrastructure provider (formerly Allbirds) that offers managed, single-tenant compute deployments for organizations requiring direct control over servers running their AI models, with a focus on data sovereignty and bespoke model hosting.

A European compute company that provides AI infrastructure services to enterprises, particularly those in industries like pharmaceuticals, energy, and finance that prioritize data sovereignty.

A distributed AI computing unit by Span, built with Dell hardware and Nvidia Blackwell GPUs, designed to be installed outside homes and small businesses to create a decentralized cloud computing network for AI workloads.

A parameter-efficient fine-tuning technique that is a variant of LoRA, used for fine-tuning models with low-rank adaptations.

A parameter-efficient fine-tuning technique that serves as an alternative to LoRA, representing approximately 3.7% of PEFT usage in code repositories.

A parameter-efficient fine-tuning technique used primarily in image generation, offering an alternative approach to LoRA for adapting model weights.

An adaptive variant of LoRA that dynamically allocates parameter budgets during fine-tuning, representing approximately 3.5% of PEFT usage in code repositories.

A benchmark and research framework by ServiceNow that evaluates privacy leakage in deep-research AI agents, containing 1,001 multi-hop research chains to test whether agents inadvertently leak sensitive information through their web queries.

A benchmark framework for evaluating deep research agents on enterprise document tasks, providing a harness for testing agent performance on multi-hop question answering.

A web document corpus used for benchmarking AI agents on web browsing and retrieval tasks, providing controlled web documents for research evaluation.

An AI-powered site reliability engineering (SRE) startup that uses AI to automatically catch and resolve bugs in software, enabling real-time monitoring and system failure resolution.

A compound AI system by OpenRouter that fans out prompts to a panel of models in parallel, then uses a judge model to synthesize the best response from multiple outputs, achieving near-frontier model performance at reduced cost.

An open-source evaluation workbench by Allen Institute for AI (AI2) designed for the LLM development loop, enabling developers to add, configure, and run benchmarks across model checkpoints with support for agentic and multi-turn evaluation, detailed per-prompt analysis, and statistical significance testing.

The Open Language Model Evaluation Standard by Allen Institute for AI, introduced in 2024 to standardize LLM benchmark scoring by pinning down prompt formatting and task formulation choices, making benchmark results more reproducible and comparable across model releases.

An autonomous AI agent developed by Andon Labs that operates a retail store in San Francisco, handling staffing decisions, inventory management, marketing, and customer interactions without human oversight.

A protocol that provides a standardized way for AI agents to communicate with and call other agents, enabling multi-agent collaboration and orchestration.

A subscription service by Google starting at $10/month that unlocks advanced Gemini-powered features for Google Home devices, including Gemini Live conversations, Nest camera queries, and Home Brief summaries.

NVIDIA's GPU-accelerated library for basic linear algebra subroutines (BLAS), providing highly optimized implementations of matrix operations including GEMM kernels used extensively in deep learning.

A free online tool by Deezer that scans playlists from various music streaming platforms (including Spotify, Apple Music, SoundCloud, and YouTube Music) to identify AI-generated tracks. It supports 27 languages and works with 20 popular streaming services.

A mobile app that uses AI to organize and categorize screenshots from your phone's camera roll into searchable collections, tracking down original links and helping users rediscover and act on saved content like recipes, products, and inspiration.

An AI chatbot integrated into the DoorDash app that allows users to order food and groceries using natural language text prompts and photos, enabling conversational search for restaurants, grocery items, and reservations.

A presentation generation feature within GenSpark that creates polished slide decks from natural language prompts, including visuals and structured content.

A voice-to-text productivity feature within GenSpark that converts rough spoken notes into structured, usable outputs like emails and meeting recaps.

An AI chipmaker that designs wafer-scale processors optimized for AI training and inference, offering an alternative to GPU-based computing for large-scale AI workloads.

A market intelligence platform that helps companies track competitors, manage sales relationships, and gather competitive insights using AI-driven analysis.

An enterprise-grade content management platform owned by Automattic that offers an enterprise version of WordPress, with capabilities focused on AI discoverability, content optimization, and helping brands maintain visibility in AI search engines.

An AI-powered feature built into Lenovo tablets that assists users with reading and content comprehension tasks.

An AI-powered note-taking feature on Lenovo tablets that provides intelligent assistance for organizing and managing notes.

A parental control tool by Google that allows caregivers to manage a child's Google account, monitor screen time activity, and set content restrictions across devices.

Snap's development platform for creating augmented reality experiences and lenses for Snapchat and Snap's AR devices, now expanded to support native development for the new Specs AR glasses.

An AI attribution platform that creates 'AI DNA' for songs, breaking them down into component parts to trace how AI models use those elements. It offers intellectual property provenance, audit and compliance reporting, model optimization, and a name, image, and likeness (NIL) attribution suite to track how artist voices, likenesses, and performance identities are used in AI training and generation.

A New York-based enterprise AI platform that connects to companies' knowledge sources via APIs to build a 'context graph' about their business, enabling AI agents to work more effectively by understanding relationships between entities, data, permissions, domain knowledge, workflows, and company-specific terminology.

An AI coding cloud startup founded by former Datadog engineers that routes between multiple AI coding models and open source options, offering model-independent infrastructure to reduce lock-in to any single AI vendor.

A Reddit-style app by Meta that includes an AI-powered 'Ask' tab, allowing users to pose questions and receive answers synthesized from discussions happening across Facebook Groups.

An AI venture founded by Jeff Bezos that raised $12 billion to build what it describes as an 'artificial general engineer' for the physical world.

Microsoft's identity and access management platform (formerly Azure Active Directory) that provides authentication and authorization services for enterprises, with emerging support for AI agent identities.

A harness developed by ServiceNow for evaluating voice models, including benchmarking ASR systems on code-switched speech across multiple language pairs.

An analytics platform that benchmarks AI models on complex, long-running analytical tasks, reporting that Claude Fable 5 was the first model to achieve a 90% score on its core analytics benchmark.

An AI-powered workspace and agent platform that evaluates and deploys AI models for tasks like UI design and game coding, reporting that Claude Fable 5 outperformed all other models in its evaluations.

An AI-powered legal platform focused on in-house legal teams at small and mid-sized businesses, providing workflow automation, intake triage, and task routing across channels like Slack, email, and Jira.

An AI-powered Safari feature by Apple that monitors changes across websites and sends notifications to users when specified conditions are met, such as product availability or content updates.

An AI-powered obstacle recognition technology built into iRobot Roomba robot vacuums that identifies and avoids common household objects like cords and socks during cleaning.

A U.S. cybersecurity platform that provides AI-powered security awareness training, including deepfake detection and phishing prevention, for employees and enterprise clients.

Apple's visual-scripting and automation app for iOS, updated in iOS 27 to allow users to build multi-app workflows by describing what they want in natural language, powered by Apple Intelligence.

Apple's cloud AI infrastructure service that processes AI requests with strong privacy protections, now offered at no cloud API cost to developers with fewer than 2 million first-time App Store downloads.

Amazon's print-on-demand service that now features AI-powered design generation through Alexa, allowing users to create custom merchandise like apparel and tumblers from text prompts, with production and Prime shipping handled by Amazon.

An Apple Intelligence feature in Safari that allows users to generate custom browser extensions by describing their desired functionality in natural language, directly from the browser's toolbar.

An AI-powered feature in Apple's Safari browser that monitors web pages for changes such as product restocks, price drops, or content updates, and sends push notifications when changes are detected.

An AI-powered PDF editor for Windows and Mac that offers editing, OCR, format conversion, annotation, batch processing, and AI features like document summarization, translation, rewriting, and chat-with-PDF functionality. It includes 20GB of cloud storage.

An Apple on-device AI feature that detects and blurs nudity, gore, and violent content in shared images and videos within Messages, designed to protect children from harmful content.

An AI-powered inventory management system that used cameras and computer vision to automatically count store items. It was deployed across 11,000 Starbucks stores before being discontinued due to inaccurate counts in real-world conditions.

An Indian AI venture platform focused on conversational AI and enterprise automation, founded by Aakrit Vaish.

An AI-powered enterprise service management platform co-founded by Vijay Rayapati, with teams spanning the U.S. and India.

An AI agent platform that functions as a cloud workforce rather than a traditional chatbot. It provides access to multiple AI models (including Claude Opus 4, GPT-5.5, and open-source models) and can generate documents, slides, and deep research reports with professional formatting, all without requiring technical setup.

An open framework for evaluating AI agents inside containerized, sandboxed environments, primarily aimed at running and publishing agent benchmarks in sealed, reproducible containers.

Google's cloud-based document editing platform, which now includes integrated Gemini AI features such as writing assistance pop-ups and smart features that can be toggled on or off through workspace settings.

A PyTorch feature that compiles and optimizes model execution by fusing operations and reducing overhead, designed to accelerate deep learning workloads without requiring code changes beyond a single function call.

An AI-powered bookmarking and organization tool that helps users save and organize links, images, and other content for easy retrieval later.

An AI-powered cart assistant within the Uber Eats app that helps users with food ordering through intelligent suggestions and conversational interactions.

An AI-powered shopping assistant by Instacart that grocers can offer to their customers to help with grocery shopping recommendations and cart building.

Google's AI watermarking tool that embeds imperceptible identifiers into AI-generated content, now integrated into Chrome and Search to help users detect AI-generated material.

A platform that tracks, measures, and optimizes the costs and performance of generative AI investments, helping enterprises manage token spending.

A multi-agent economic simulation built on a 3-billion-parameter language model, where five AI agents trade goods, react to market shocks, and produce emergent economic behaviors in real time.

Snowflake's AI service that provides machine learning and AI capabilities built into the Snowflake data platform, enabling users to build and deploy AI-powered applications on their data.

OpenAI's program similar to Anthropic's Cyber Verification Program, designed to provide cybersecurity professionals with expanded access to AI capabilities for security research and work.

An AI cybersecurity startup focused on applying artificial intelligence to cybersecurity challenges and defense.

An AI development platform founded by Ankur Goyal that provides tools for evaluating, testing, and monitoring AI applications and large language models.

A benchmark for evaluating AI models on real-world software engineering tasks, measuring how well AI coding agents can resolve actual GitHub issues in production codebases.

A new AI coding benchmark designed to measure advanced code generation capabilities, where even top frontier models score relatively low, indicating it tests cutting-edge programming ability.

A cybersecurity benchmark that evaluates AI models on their ability to identify and reason about software vulnerabilities and exploits.

Anthropic's early access program that provides select companies and partners with access to pre-release models, including the Mythos class of Claude models, before their public launch.

The official command-line interface for the Hugging Face Hub, enabling users and AI agents to download/upload models, datasets, and Spaces, manage repos, run jobs, and interact with Hub infrastructure directly from the terminal.

An open-source benchmark by ServiceNow AI for evaluating voice agents across enterprise domains, featuring 213 scenarios across 121 tools spanning airline customer service, IT service management, and healthcare HR service delivery.

An AI-powered conversational assistant built into Facebook that provides creators with personalized recommendations, performance insights, content brainstorming ideas, and answers to questions about their audience and posting strategy.

A privacy-focused search engine that has made its 'no-AI' search experience easier to access, experiencing a traffic boom as users seek alternatives to AI-integrated search engines.

A Substack moderation feature that uses AI to automatically learn from a writer's moderation decisions and hide comments that violate custom natural-language rules set by the writer.

A speech-to-text benchmarking tool by Pipecat used to evaluate semantic word error rates in ASR systems.

Apple's device-tracking service that uses a crowdsourced network of Apple devices to help users locate lost or stolen Apple products, accessories, and compatible third-party items.

A Google Labs app for iOS and Android that uses data from a user's connected Google services (Gmail, Calendar, Photos, YouTube, Search History) to generate a curated daily collection of AI-illustrated lifestyle stories, recommendations, and inspiration.

Coralogix's AI agent designed to help engineers investigate incidents and query operational data within the Coralogix observability platform.

A desktop application by Nous Research that provides a local AI agent interface supporting multiple LLM providers, API key management, memory/context persistence, voice chat, and integration with messaging platforms like Discord. It can connect to remote Hermes backends and supports various models including DeepSeek.

A lightweight Python library that enables efficient quantization of large language models, including 4-bit NF4 quantization, to reduce memory usage during inference and training.

A reinforcement learning research project from UC Berkeley focused on training and evaluating AI agents, contributing to the open-source agentic RL ecosystem.

An open-source tool designed to streamline the fine-tuning of large language models, supporting various training configurations and methods including LoRA, QLoRA, and full fine-tuning.

An AI-driven IT helpdesk startup that uses artificial intelligence to automate IT support and troubleshooting tasks for enterprises.

A startup that uses AI to simulate user behavior for market research, enabling companies to test and analyze consumer responses without traditional focus groups.

A proposed Apple intermediary framework designed to allow third-party virtual assistants to safely access the same features and capabilities as Siri AI on devices in the EU, while protecting user privacy.

A toolkit by NVIDIA for optimizing and quantizing AI models for efficient inference, supporting configurations like W4A16 NVFP4 for faster deployment on NVIDIA hardware.

An open-source standard by Microsoft that provides developers, compliance, and security teams a consistent way to define and enforce policies governing AI agent behavior, including what agents may do, what they must not do, and when human approval is required.

An AI compliance service that sits between AI models and end users to flag and rewrite messages that may present compliance problems, using deterministic rules for standards like SOC 2 and GDPR combined with LLM-powered rewrites.

An open source framework by Microsoft (Adaptive Spec-driven Scoring for Evaluation and Regression Testing) that turns natural-language descriptions of intended AI behavior into structured, scored test cases for application-specific AI evaluation and regression testing.

An always-on agentic AI assistant by Microsoft built on the OpenClaw framework, designed to work within the Microsoft 365 ecosystem with a persistent identity that actively adapts to user behavior over time. It offers prepackaged skills for calendar management and meeting agendas, and allows users to develop custom skills through ongoing feedback.

Google's official phone dialer app for Android that now includes AI-powered fake call detection to protect users against deepfake voice impersonation scams by verifying caller identity through a device-to-device confirmation system over RCS.

Microsoft's Android-based software platform described as a chip-to-cloud system designed for securely building, deploying, and running multiple AI agents in an open, multi-agent environment.

An AI-powered productivity feature integrated into Notion's workspace platform, leveraging models from providers like Anthropic to assist with writing, summarization, and automated workflows.

An IBM AI-powered tool designed to accelerate mainframe application development and modernization, featuring an App Insights agent that leverages deep static analysis for application understanding of legacy code written in COBOL and PL/1.

An IBM proprietary program analysis and data processing library used for agent-based generation of unit, integration, API, and change-based tests, achieving higher developer ratings and superior code coverage benchmarks compared to open-source tools and zero-shot LLMs.

An AI-powered weather forecasting model developed by WindBorne Systems that uses deep learning and proprietary balloon sensor data to produce hourly forecasts with resolution down to 3 km, claiming greater accuracy than traditional forecasting systems including those from the ECMWF.

A $100 screenless wearable fitness tracker by Fitbit (Google) that features AI health coaching, an 8-day battery life, and focuses on pure health tracking without notifications or screen distractions.

An AI-powered personal context portfolio builder available at contextportfolio.ai that helps users create and manage their personal context for AI interactions.

A cloud platform for running serverless GPU workloads, enabling developers to deploy and scale AI models and compute-intensive applications without managing infrastructure.

A data center tracking platform founded by Michael Thomas that monitors and reports on the construction and operation of AI data centers across the United States.

An AI-powered web browser by Perplexity that acts as a chatbot-based search engine, capable of summarizing emails, browsing web pages, and performing tasks like sending calendar invites.

An AI agentic browser by Opera with contextual awareness that can perform tasks like researching, shopping, and writing code snippets, even while the user is offline.

A Y Combinator-backed AI-first, browser-native automation platform designed to autonomously complete tasks, fill out forms, and manage data across services like Gmail, Notion, Slack, Figma, and banking platforms by operating directly within the browser.

A privacy-focused web browser from the DuckDuckGo search engine company that blocks trackers and ads, doesn't track user data, features an enhanced scam blocker, and has recently introduced generative AI features including a chatbot.

A platform offering a free Claude-based AI skill and organizational transformation framework that allows businesses to run their operations using an exponential organization model.

A service by Hugging Face that allows users to deploy machine learning models as production-ready API endpoints on managed infrastructure.

An API developed by Thinking Machines Lab for fine-tuning open-source AI models, enabling developers to customize models for specific use cases.

A personal AI assistant launched by Microsoft, inspired by OpenClaw, designed to help users with everyday tasks.

AI-powered smart glasses developed in partnership between Meta and Ray-Ban, featuring cameras and AI assistants that can see and respond to the wearer's environment in real time.

A built-in profiling module in PyTorch that provides statistical summaries and temporal execution traces of CPU and GPU activities, helping developers identify performance bottlenecks in model training and inference.

A smart bird feeder with a built-in 4K AI camera that uses a proprietary bird-identification algorithm to identify over 10,000 bird species, sending notifications and recording visits via a companion app.

DJI's companion app for its drones that provides video editing tools including trimming, keyframing, filters, and an AI-powered in-app tracking feature that automatically follows selected subjects in 360-degree footage.

A coding benchmark developed by DataCurve that evaluates AI models on realistic, novel software engineering tasks built from scratch, designed to avoid memorization issues and better reflect real-world long-horizon coding capabilities.

OpenAI's autonomous coding agent that can execute software engineering tasks independently, designed for professional development workflows.

A platform that lets AI agents build and work with disposable copies of databases, enabling safe experimentation without risking production data. It provides isolated database environments for each AI agent to test changes independently.

A custom-built benchmark and game simulation designed to test how well different large language models can learn iteratively by writing scripts to control ships navigating gravitational physics challenges over 30 rounds of feedback.

A cloud security platform acquired by Google for $32 billion that identifies and remediates security vulnerabilities in real time, including in AI-generated code.

A personal AI assistant launched by Microsoft, inspired by the OpenClaw project, designed to help users with everyday tasks and information retrieval.

A Y Combinator-backed AI startup focused on machine learning tools and services.

An AI-powered feature for YouTube Premium users that generates personalized radio stations, playlists, and podcast recommendations based on genres, mood, or listening history.

An inference neocloud startup that rents out AI processing power optimized for the inference phase, deploying specialized SambaNova chips to deliver high-speed token generation for AI workloads.

An Intel-backed AI chipmaker that designs specialized processors optimized for AI inference workloads, claiming superior performance over GPUs and competing inference chips from Groq and Cerebras.

A fully managed search and vector database service by AWS designed for agentic AI workloads, capable of instantly scaling compute up during agent traffic bursts and scaling down to zero when idle, decoupling compute from storage to optimize cost.

An agent builder within Asana's work management platform that allows users to create AI-powered automations and deploy AI teammates to handle business workflows and processes.

Cloudflare's infrastructure platform providing persistent environments and instant scalability for AI agents, designed to handle machine-generated traffic patterns and agentic workloads.

A conversational AI platform founded by Oculus co-founders that offers distinct AI agents (Maya, Miles, Simone, and Charlie) with unique voices, personalities, and memory, designed to deliver natural-feeling spoken conversations with real-time search integration.

An AI research project by Andrej Karpathy that uses agent swarms to train LLMs on simple tasks, aiming toward recursive self-improvement of language models. Building blocks are available via a public GitHub repository.

A software development kit by Anthropic for building and deploying AI agents powered by Claude models, enabling tool use and multi-step task execution.

An open-source framework by Microsoft for building multi-agent AI systems where multiple AI agents can converse and collaborate to solve complex tasks.

An open-source SDK by Microsoft that integrates large language models into applications, enabling AI orchestration with plugins, planners, and memory capabilities.

A set of .NET libraries by Microsoft that provide unified abstractions for integrating AI services into .NET applications, offering a consistent API across different AI providers.

An AI-powered platform for building web applications from natural language prompts, enabling rapid prototyping and deployment.

A trust and safety company (now rebranded) that provides content moderation and online safety assessment tools, used by platforms like Meta to stress test and evaluate teen safety features.

A benchmark and evaluation framework by MLCommons designed to measure how AI models behave under different conditions, focusing on safety and responsible AI assessment.

A holistic evaluation framework developed by Stanford's Center for Research on Foundation Models (CRFM) that provides comprehensive benchmarks to assess language model capabilities across multiple dimensions.

Google's e-book platform that now includes AI-powered features such as 'Catch me up' for story recaps and the ability to highlight passages to ask questions about the text.

The core AI intelligence layer behind Microsoft 365 Copilot, providing the underlying AI capabilities that power Microsoft's productivity agent experiences.

Microsoft's scientific research platform powered by AI, now generally available, designed to accelerate scientific discovery and research workflows.

A sandboxing system by Microsoft that provides enterprise-grade sandbox environments for AI agents, with OS-enforced containment to securely run agent code, file access, and network interactions.

A benchmark developed by Artificial Analysis and IBM that evaluates AI models on agentic enterprise IT tasks, starting with Site Reliability Engineering (SRE) tasks involving Kubernetes incident response, diagnosis, and root-cause identification.

A lightweight, accurate voice activity detection model that runs on CPU, widely used as the de-facto default VAD in open-source voice agent pipelines.

Snowflake's AI building tool that enables enterprise users to interact with their data using natural language queries, generate summary reports, and build AI-powered features on top of Snowflake's data platform.

An autonomous AI software engineer developed by Cognition that can independently plan, write, debug, and deploy code, serving enterprise customers like Mercedes-Benz, NASA, and Goldman Sachs.

A closed-loop, in-ear sleep system by SOND that captures 12 physiological signals in real time and uses a cloud-based AI sleep coach to select or generate personalized sleep audio programs to help users improve their sleep.

An open-source columnar database management system designed for real-time analytics on large datasets, offering managed cloud services particularly suited for processing the massive data requirements of AI agents.

A beta feature from Robinhood that enables users to create dedicated accounts for AI agents, allowing those agents to read portfolios, analyze data, suggest investments, and execute stock trades using a pre-loaded wallet balance.

Samsung's AI-powered TV processor that uses 128 neural networks to automatically upscale non-4K content to higher resolution and reduce motion blur on Samsung Neo QLED displays.

DuckDuckGo's dedicated AI-free search page that turns off all AI features including AI-assisted answers and AI-generated images by default.

A Silicon Valley-based startup that collects egocentric video and multi-sensor data from gig economy workers using custom hardware devices to create training datasets for robotics and physical AI models.

A fine-tuned version of Gemini developed by Google DeepMind with specialized tools for scientific research, including hypothesis generation, data analysis, and literature summarization, functioning as an AI research assistant.

A startup that uses AI automation to handle all software operations for solopreneurs, reportedly run by a single person and valued at $250 million.

An AI-powered cryptocurrency analysis platform that provides real-time market insights, trading indicators, structured strategies, and expert analysis to help traders make smarter investment decisions. It includes a Discord community with daily market analysis and altcoin coverage.

A Hugging Face Spaces project that uses AI to generate simple HTML creations like clocks, to-do lists, and basic games in one shot, inspired by The Amazing Digital Circus animated show.

An AI-powered wearable pendant under development by Meta, building on technology from its acquisition of Limitless, designed to be worn as a necklace or clipped to clothing for conversational AI assistance.

A wearable AI device that clips onto clothing and uses an on-board microphone to record conversations throughout the day, generating AI-powered summaries of interactions and meetings.

A reliability engineering agent startup that analyzes AI-generated code quality, reporting that companies spend 44% of their tokens on fixing bugs introduced by AI-generated code.

An AI-powered legal assistant by Thomson Reuters that helps lawyers with legal research, document review, and other routine legal tasks.

IBM's enterprise AI and data platform used to process and analyze large volumes of data, power AI-driven fan engagement, content personalization, and storytelling applications for businesses and sports organizations.

Visa's suite of AI-powered payment tools designed to enable agentic commerce, allowing AI agents and developers to process and manage payments securely.

A tracking service that monitors daily GPU rental pricing across 28 marketplaces and cloud providers, providing market data on GPU costs for AI compute.

A pair of specialized small language models by Dharma AI designed for structured OCR tasks, particularly Brazilian Portuguese document extraction. A 3-billion-parameter model that outperformed major frontier APIs on domain-specific OCR benchmarks at roughly fifty times lower cost.

Google's personal AI agent that runs 24/7 on Google's virtual machines and can take actions across Gmail, Docs, Sheets, and other Google Workspace apps on behalf of the user. It supports MCP (Model Context Protocol) for connecting third-party apps and is available to Gemini Ultra plan subscribers in the US.

A wearable AI hardware device by Humane that was designed to serve as a screenless, voice-and-gesture-controlled AI assistant worn on clothing. It launched to poor reviews and commercial failure.

A wearable AI companion device worn as a necklace that listens to conversations and provides AI-powered interactions and companionship throughout the day.

An open-source reference harness for running agentic AI evaluations, providing shell access to sandboxed file systems for models to investigate and solve tasks with a configurable turn cap.

Hugging Face's managed service for deploying machine learning models as production-ready API endpoints, supporting various model architectures and scaling configurations.

A $7.99/month subscription plan from Meta that provides enhanced AI capabilities including deeper reasoning for complex tasks and expanded image and video generation features across Meta's apps.

A $19.99/month subscription plan from Meta offering the highest tier of AI compute, including extended thinking mode, deeper reasoning for complex tasks, and expanded video and image generation capabilities.

Google's AI-powered coding agent designed to assist developers with software engineering tasks, bolstered by Google's acqui-hire of the Windsurf team.

An AI-powered investment research platform acquired by Robinhood in 2024 that provides data-driven analysis and insights for investors.

A startup building payment products that give AI agents the ability to make purchases and transactions on behalf of users.

A standalone desktop app by Spotify that uses AI to generate personalized podcasts by connecting to users' email, calendar, and other personal data sources, with an agent that can browse the web and fetch information.

An AI therapy and coaching app co-founded by Tony Robbins and former Calm employees that offers personalized mental health support through specially trained AI models and virtual AI therapists, scoring 95 on mental health safety benchmarks.

Google's AI-powered phone calling service that can make real-world bookings at restaurants and salons on behalf of users, first demonstrated at Google I/O 2018.

An AI-powered waste sorting system that uses sensors, visible light, and infrared cameras combined with robotic arms to identify and separate recyclable materials including aluminum, plastics, and other waste stream components with over 90% accuracy.

A metals recycling startup that uses AI algorithms combined with lasers, cameras, and X-ray fluorescence sensors to classify and sort aluminum scrap by specific grade, enabling higher-accuracy separation and greater profit per pound.

A paid, ad-free search engine that offers customizable search experiences with features like website filtering, search lenses, and an optional AI-powered 'Quick Answer' summarization feature that can be toggled off.

A privacy-focused search engine that does not collect user data such as search, browsing, or purchase history, and offers optional AI-generated answers that users can disable in settings.

A privacy-focused search engine that acts as a proxy for Google, stripping personal data like IP addresses from queries before sending them to Google and returning results anonymously.

A lightweight search tool that automatically appends the &udm=14 parameter to Google searches, delivering standard Google results without AI Overviews or other AI-generated content.

A Spotify Premium feature that allows mobile users to ask AI-powered questions about podcast episodes they are listening to, get answers about concepts mentioned, and receive podcast recommendations.

A marketing company that deploys AI agents to mass-publish content across Reddit and blogs to influence Google and ChatGPT rankings through generative AI-engine optimization (GEO).

An AI-powered desktop companion for PCs and Macs that learns a user's daily workflows and automates them with limited to no human prompting. It includes a built-in skills library for tasks like email drafting, invoice processing, report building, and document summarization, as well as a coding assistant.

An Andreessen Horowitz-backed AI search startup that provides an AI-native search API and platform, valued at $2.2 billion after raising $250 million to transform how search and discoverability work.

An AI assistant built into Figma's collaborative design canvas that uses natural language prompts to generate new designs, edit existing ones, and automate design tasks, powered by AI models fine-tuned for design use.

A marketing infrastructure platform that uses AI to optimize the distribution strategy and clipping process for short-form video content, leveraging a network of over 100,000 gig creators and continuous testing loops to determine what makes content go viral.

Google's in-car platform that integrates with vehicle infotainment systems, now incorporating Gemini-powered AI features to provide more helpful and safer in-car interactions for drivers.

A command-line AI agent tool that supports plugging in any model, offering a flexible and model-agnostic approach to agentic coding workflows.

A collaboration and project management software platform that has integrated approximately 3,000 internal AI agents to automate complex tasks and boost employee productivity.

A platform by Allen Institute for AI that enables partners and organizations to deploy OlmoEarth models for satellite imagery analysis and environmental monitoring at scale.

A Hugging Face skill for AI coding agents that bootstraps training recipes for Sentence Transformers models, enabling fine-tuning of cross-encoders and embedding models through AI-assisted development.

Google's command-line interface tool (now stable at version 1.0) that enables AI coding agents to access Android development knowledge and capabilities from Android Studio, regardless of the user's preferred coding platform.

A Google shopping hub that lets users add products from across Google services (Search, Gemini, YouTube, Gmail), track deals, monitor price drops, surface price history, and flag product compatibility issues using AI.

A Google protocol designed to prevent AI agents from making unauthorized purchases by allowing users to set strict spending limits, purchase restrictions, and approved merchant lists for AI agent transactions.

A Google protocol that enables AI agents to purchase products, order goods, and reserve hotels on behalf of users, with partners including Shopify, Amazon, Walmart, Stripe, Salesforce, and Meta.

A Gemini AI-powered conversational feature for Gmail that allows users to ask natural language questions about their inbox using voice, find buried information, handle follow-up questions, and navigate between topics seamlessly.

An AI-powered Gmail feature that provides an overview of tasks and items to catch up on from a user's inbox, organized on a single page, expanding from Google AI Ultra to Pro and Plus subscribers.

AI agents within Google Search that operate continuously in the background to monitor, synthesize, and deliver updates on user-specified topics of interest, functioning as an advanced evolution of Google Alerts with agentic capabilities.

An agentic email security platform that uses AI to analyze the context of every incoming email to detect fraud, phishing, and impersonation attempts, particularly those powered by AI. It employs a small language model tailored to quickly analyze emails, understand sender intent, and evaluate against organizational context.

Google Maps' panoramic street-level imagery service covering 110 countries with over 280 billion images, now integrated with Google's Genie world model to enable interactive, AI-generated simulations of real-world locations.

Google's AI-powered search experience that integrates Gemini 3.5 Flash to provide AI-generated answers and agentic capabilities directly within Google Search.

An AI-powered conversational search feature within Google Search that has surpassed 1 billion monthly users, allowing users to ask natural language questions and have back-and-forth conversations about search results.

Google's agentic development platform that enables the creation of AI agents capable of performing tasks autonomously, used in combination with Gemini to power interactive search experiences.

Sony's AI-powered audio upscaling technology that enhances compressed audio files to near high-resolution quality, providing an improved listening experience on compatible Sony headphones and devices.

Samsung's AI-powered motion processing technology built into its TVs that smooths fast on-screen movements and reduces blur or flickering, particularly useful for sports viewing and fast-paced content.

A voice-enabled AI feature for Google Docs that allows users to dictate their thoughts and have AI brainstorm, write, outline, and refine content into a first draft.

A Gmail feature that acts as a personalized briefing tool, providing updates on email topics, suggesting next steps, generating draft replies, offering instant access to relevant Google Docs/Sheets/Slides, and streamlining task management.

A YouTube app for video creation that is integrating Gemini Omni Flash to enable AI-powered video generation and editing for YouTube Shorts, available at no cost to users.

Apple's gesture-based screen reader enhanced with AI capabilities including Image Explorer for detailed content descriptions and Live Recognition activated via the iPhone Action button for follow-up questions about on-screen content.

An Apple Intelligence-powered feature within VoiceOver that provides detailed AI-generated descriptions of device displays and content for users who are blind or have low vision.

An Apple accessibility tool enhanced with AI that helps users navigate complex text like scientific studies with columns, images, and tables, providing on-demand summaries and language translation while preserving custom formatting.

An opt-in AI-driven service from Google that aggregates information from a user's email, calendar, and other sources to generate a daily summary with prioritized tasks and upcoming events, powered by Gemini AI.

A new Google feature that integrates YouTube video results directly into Google Search, allowing users to ask conversational questions and watch relevant YouTube tutorials or videos within an AI-generated interactive search response.

An AI-powered wearable device worn on the wrist that records, transcribes, and summarizes the user's conversations throughout the day, functioning as a personal assistant with calendar integration and reminder features.

An open framework for running and reproducing AI agent evaluations, designed to work alongside the Open Agent Leaderboard for standardized agent benchmarking.

An OCR and document parsing toolkit by PaddlePaddle that supports multiple inference backends including Hugging Face Transformers, providing capabilities for text recognition, document layout analysis, and structured data extraction from documents.

A feature within Amazon's Alexa+ that generates on-demand podcast episodes by researching a user-specified topic, creating a script, and narrating it with AI-generated host voices, with options to customize length, tone, and focus.

A Swiss deep tech company spun out of ETH Zurich that builds an AI-powered augmented reality motorcycle helmet displaying navigation, speed, and safety alerts anchored to the real-world road ahead using LetinAR's optical modules.

An AI-powered website builder that creates complete websites and online stores from a simple text description of your business, including branded pages, navigation, product layouts, logos, and product descriptions — all without requiring any coding knowledge.

An internal Meta tool that tracks U.S. employees' mouse movements, keystrokes, clicks, and other computer interactions to build AI agents capable of performing software tasks autonomously.

An AI-native database platform at ghost.build designed specifically for AI agents rather than human users, featuring no traditional dashboard or UI and built to be accessed programmatically by autonomous agents.

Google Cloud's computer vision API that provides image analysis capabilities including OCR, object detection, and label detection for developers and enterprises.

Google Cloud's document processing platform that uses machine learning to extract structured data from unstructured documents, including forms, invoices, and other business documents.

An AWS machine learning service that automatically extracts text, handwriting, and structured data from scanned documents, going beyond simple OCR to identify form fields and table contents.

Google's AI-powered translation service that provides real-time language translation, integrated into Android XR glasses for live translation capabilities.

An AI-powered legal technology startup that provides AI tools designed to assist with legal document drafting and review.

A legal AI startup that offers AI-powered tools for legal professionals to streamline document-related workflows.

The latest version of Replit's AI-powered coding agent that enables vibe coding — building apps through natural language prompts. Agent 4 introduces parallel agents for working on multiple ideas simultaneously, collaboration features for merging project flows, and multi-workspace project views.

Google's AI-powered reinvention of Google Alerts that deploys background AI agents to monitor topics 24/7, helping users track market trends, price changes, weather warnings, and other interests.

A Gemini-powered AI feature that compiles a personalized daily digest from a user's Gmail inbox, calendar, and tasks, delivering a summarized update of their day.

An app built by former NotebookLM developers that enables users to create personal AI-generated podcasts on any topic.

A new app by Meta/Facebook designed for deeper discussions and Q&A, leveraging Facebook Groups communities to provide real answers, with an AI assistant that helps fetch answers and assists group admins with moderation.

An open source, Apple-only LLM server that lets Mac users run and switch between local and cloud AI models while keeping files, tools, and model memory on their own hardware in a sandboxed environment.

Google's health and fitness companion app that replaced the Fitbit app, featuring AI-powered coaching via Gemini integration, health data tracking, and personalized fitness insights for Fitbit device users.

An AI search startup focused on building next-generation search and discoverability tools as part of the emerging wave of AI-native search platforms.

A node-based design tool that was acquired by Figma to enhance its design capabilities and product offerings.

A startup operating in the automated video clipping space, competing with platforms like Clouted to help brands create and distribute short-form video content.

An enterprise marketing infrastructure platform focused on influencer and creator marketing management, used by brands to manage and measure creator-driven campaigns.

Google's upcoming AI-powered extended reality smart glasses running on the Android XR platform, designed to overlay intelligent features onto the real world.

Google's core search engine, which is being fundamentally transformed with generative AI capabilities to provide AI-generated overviews and more conversational search experiences.

An open source hardware project that pairs a Waveshare ESP32-S3-Touch-AMOLED-2.16 device with your laptop over Bluetooth to display Claude Code token usage statistics, session data, and pixel-art animations on a tiny desktop dashboard.

A data supply platform that connects over 700,000 artists and designers with AI labs, providing multimodal datasets of images, videos, design assets, and 3D content for training foundation models.

An AI-assisted coding feature built into Microsoft Visual Studio that provides predictive suggestions, context-aware recommendations, and intelligent code completion to help developers write code more efficiently.

Google's integration of Gemini AI directly into Android Studio IDE, providing AI-powered coding assistance specifically tailored for Android app development.

Google's official integrated development environment (IDE) for Android app development, now enhanced with AI capabilities including Gemini integration for code assistance.

Google's modern declarative UI toolkit for building native Android apps using Kotlin, now integrated into Google AI Studio's app creation workflow.

Google's AI creative studio that integrates with Gemini Omni Flash for video creation and other creative workflows.

A new Android system feature by Google that allows users to track the progress of AI agents like Gemini Spark on their mobile devices.

A startup founded by Andrej Karpathy dedicated to applying AI assistants to education, aiming to transform how people learn through AI-powered tools.

Google's note-taking and to-do list application that is gaining voice-powered AI capabilities similar to Gmail Live for conversational interaction with notes and tasks.

A Google Search feature that connects with personal data sources like Gmail to provide personalized AI-powered search results, expanding to nearly 200 countries and 98 languages.

A Google Labs experimental feature that sends users email summaries including 'top of mind' and 'FYI' sections along with a calendar summary, serving as the precursor to Google's Daily Brief feature.

A new suite of AI services from Anthropic designed specifically for small business owners, offering automated bookkeeping, business insights, ad campaign generation, and integrations with tools like QuickBooks, Canva, Docusign, HubSpot, and PayPal.

A startup that serves as a marketplace connecting video game companies with AI world-model labs, enabling the licensing and conversion of video game assets into high-quality training data for AI models that need to understand physical-world dynamics.

Amazon's personalized AI shopping assistant powered by Alexa+ that replaces the earlier Rufus assistant, offering voice- and touch-enabled shopping with personalized recommendations, price tracking, recurring orders, and cross-retailer purchasing capabilities.

Anthropic's law-focused plug-in for Claude that provides legal-specific AI features tailored for lawyers and legal professionals.

Notion's new developer platform that extends custom AI agent capabilities, connects with external agents, and allows teams to build automated multistep workflows with custom code deployment and database syncing.

An AI-powered productivity and collaboration workspace that combines features similar to Notion and Miro, offering AI-assisted summarization, rewriting, idea refinement, document creation, task planning, and visual workflow building in a single platform.

An AI-powered robotic pool cleaner by Beatbot featuring advanced navigation, automation, and high-performance all-zone cleaning coverage across floors, walls, waterline, and surface.

Google's AI-powered search experience that replaces traditional blue links with an AI agent that answers queries, executes tasks, and runs background monitoring agents.

A benchmark suite for evaluating AI agents on customer service and technical support tasks, including airline and retail scenarios that require following company policies.

A machine learning experiment tracking and visualization platform that helps developers log, compare, and reproduce model training runs.

A developer tools startup that automated the creation and maintenance of production-ready SDKs from API specifications across multiple programming languages, acquired by Anthropic for over $300 million.

An AI voice platform that enables companies to build, deploy, and manage voice agents for customer support, lead qualification, appointment scheduling, and outbound sales. It provides low-latency voice infrastructure and orchestration tools, and has processed over 1 billion calls.

A proactive AI personal assistant app by Second Nature Computing that consolidates calendar, email, messages, and location data into a single dashboard, using AI to anticipate user needs and offer contextual suggestions.

An AI tool by Adaption that helps models learn specific capabilities quickly by co-optimizing both data and models through an automated approach to conventional fine-tuning, enabling continuous model improvement.

A voice AI agent developed by Pair Team that serves as a 24/7 patient-facing interface for healthcare, handling intake, coordinating referrals, and conducting check-in conversations with patients between clinical visits.

A Google Android feature that lets users vibe-code custom widgets by describing what they want in natural language, with Gemini generating personalized dashboards that can pull data from the web and Google apps.

An AI feature integrated into the Google Chrome browser that allows users to summarize web page content, ask questions about what they see, and use an experimental auto-browse feature to navigate websites and complete tasks on the user's behalf.

An AI-powered mouse cursor feature by Google for Googlebook laptops that suggests contextual AI actions based on what the user hovers over, such as scheduling meetings from dates found in emails.

Google's widely used Android keyboard app that incorporates AI-powered features including predictive text, voice dictation, and the new Gemini-powered Rambler feature for intelligent speech-to-text transcription.

A production-focused AI design tool that allows teams to run and iterate on their existing codebases in the cloud, enabling designers to work directly in production environments and hand off work to developers seamlessly.

An AI-powered project management platform by YaseenAI, Inc. that supports Kanban boards, Gantt charts, sprint planning, issue tracking, and built-in documentation with AI features for generating summaries, organizing information, and automating repetitive administrative tasks.

An e-signature platform by YaseenAI, Inc. that offers unlimited document signing with AI-powered field detection via its Nova AI assistant, which automatically identifies where signatures, initials, and dates should be placed in documents.

xAI's coding and app-building tool that allows users to generate software applications using AI, released as part of the Grok platform.

A financial data connectivity platform that enables applications to connect with users' bank accounts, credit cards, and investment accounts, now integrated with ChatGPT Pro for AI-powered financial analysis.

A cyber defense suite of tools by OpenAI that leverages the company's large language models to find and remedy software vulnerabilities, offering capabilities such as secure code review, threat modeling, patch validation, malware analysis, and penetration testing.

A privacy-focused AI platform that provides access to various AI models with an emphasis on uncensored and private interactions.

A desktop application that allows users to discover, download, and run local large language models on their personal computers with a user-friendly graphical interface.

An AI feature built into the Durobo Krono e-reader that provides AI-powered assistance directly on the e-ink device.

Google's revamped Search experience that foregrounds AI-generated summaries at the top of search results, pushing traditional web links further down the page.

An AI startup founded by Ian Crosby that aims to build a fully autonomous AI bookkeeper capable of generating accrual-based financials without direct human involvement, targeting AI and software startups as customers.

An open format for representing machine learning models that enables interoperability between different ML frameworks and optimized deployment across various hardware platforms.

Intel's open-source toolkit for optimizing and deploying AI inference, particularly focused on CPU-optimized performance for deep learning models across Intel hardware.

A data labeling platform that provides high-quality human-generated datasets for training and fine-tuning AI and machine learning models.

An AI-powered search and chatbot platform founded by Richard Socher that provides conversational search and AI assistant capabilities.

An AI platform co-founded by Tim Shi that provides real-time intelligence and coaching for contact center agents, helping businesses improve customer interactions and sales performance.

OpenAI's business-tier offering of ChatGPT designed for smaller teams, providing enterprise-grade AI assistant capabilities with team management and collaboration features.

An Amazon Alexa for Shopping feature that enables the AI assistant to autonomously shop other online stores and handle purchases on behalf of the user, extending Amazon's shopping capabilities beyond its own marketplace.

A data intelligence platform for legal research that was acquired by Clio, enabling lawyers to use AI-powered legal research capabilities within the Clio ecosystem.

An OpenAI tool that uses AI capabilities to identify and find security flaws and vulnerabilities.

A conversational AI platform specializing in voice assistants for enterprise customer service, enabling natural-sounding phone interactions without human agents.

An AI voice platform that enables businesses to automate phone calls using AI-powered voice agents for sales, support, and other use cases.

An AI voice agent platform that allows developers and businesses to build, test, and deploy conversational voice AI agents for phone interactions.

A company that mass-produces AI-generated podcast episodes, contributing to the rise of automated audio content creation.

AWS accelerated computing instances featuring NVIDIA H100 GPUs designed for large-scale foundation model training and inference workloads, available in configurations including p5.48xlarge (8 GPUs) and p5.4xlarge (single GPU).

GenEditBio's AI platform that analyzes data to identify how chemical structures correlate with specific tissue targets and predicts optimal delivery vehicle chemistry for gene-editing tools.

InfiniMind's AI-powered platform that analyzes television content in real time, helping media and retail companies track product exposure, brand presence, customer sentiment, and PR impact.

A production-grade calendar management environment built by Turing for OpenEnv, serving as a benchmark for evaluating tool-using agents under realistic constraints such as access control, temporal reasoning, and multi-agent coordination.

Spotify's internal AI-powered development system that enables remote, real-time code deployment using generative AI. It integrates with tools like Claude Code and Slack to allow engineers to fix bugs and add features from their mobile devices.

An open-source OCR (Optical Character Recognition) AI model by Zhipu AI (ZAI) that can parse text, tables, formulas, handwriting, and receipts from images. At only 2.6 GB, it outperforms both open-source and closed-source alternatives in accuracy and speed.

AI coding agents capable of autonomously building entire software applications, pushing code to GitHub, and deploying to platforms like Vercel based on high-level instructions.

A Reddit-like social network designed for AI agents to communicate with one another, post, comment, and browse content. It gained viral attention when posts appeared to show AI agents organizing autonomously, though security vulnerabilities revealed that humans could easily impersonate agents on the platform.

A Swedish AI startup that helps dentists' practices with administrative work, including a recording tool that uses AI to generate clinical notes from patient visits.

Domain purchased by Crypto.com founder Kris Marszalek for $70 million, planned to offer consumers a personal AI agent for messaging, app usage, and stock trading.

xAI's project spanning from simple computer use simulation to modeling entire corporations, described as being able to do anything on a computer that a computer can do.

An AI startup founded by Anna Goldie and Azalia Mirhoseini that builds AI tools to automate and dramatically accelerate chip design, using deep learning agents that improve through experience across different chip layouts.

A web browser being developed by OpenAI as part of its expanding product suite.

An autonomous AI agent that can run locally or on a cloud VPS, capable of coding, managing Kanban boards, and completing tasks autonomously while the user is away.

An open-source OCR model for parsing text and data from images, used as a benchmark comparison for document understanding tasks.

An AI cybersecurity tool founded by AI engineer Artem Sorokin, focused on addressing security challenges in AI systems.

A Swedish legal AI startup that emerged from the SSE Business Lab incubator, applying artificial intelligence to legal workflows and processes.

A 7-billion parameter image generation model by Photoroom, trained using a carefully curated data pipeline with long captions and diverse datasets for high-quality image synthesis.

An AI creative platform offering image and video generation capabilities with access to multiple AI models for content creation.

An AI-powered photo editing application by Skylum for Windows and Mac that uses AI features like SkyAI for sky replacement, RelightAI for lighting adjustments, AI-based skin retouching, object removal, and an AI Assistant that analyzes images and recommends edits. It offers a one-time lifetime purchase instead of a subscription model.

An AI image retouching and enhancement model developed by Topaz Labs for professional-grade photo editing and restoration tasks.

An image-generation AI model created by Unconventional AI that demonstrates how oscillator-based computing architecture can replicate the performance of conventional state-of-the-art diffusion models while promising dramatically lower power consumption.

Meta's AI image generator built by Meta Superintelligence Labs, available through the Meta AI app, Instagram Stories, and WhatsApp. It supports prompt-based image creation and editing, including the ability to manipulate other users' public Instagram photos.

An open-source AI image generation model that produces highly photorealistic images with natural imperfections, supports multiple art styles including anime, and can run locally on consumer GPUs with as low as 8GB VRAM. It is notable for being uncensored out of the box and offering fast generation times.

A family of AI image generation models by Recraft, a professional AI-native design platform for generating photorealistic images, illustrations, logos, and vectors with improved natural aesthetics.

AI-powered editing features in Apple's Photos app that use generative models to reframe images by adjusting perspective, extend photos by generating new content beyond the original frame, and remove unwanted objects with improved realistic infill.

An AI-powered photography tool by Apple that allows users to change the angle, perspective, or zoom of existing photos using spatial computing models and on-device AI, filling in gaps with generative AI while preserving as much of the original image as possible.

An AI image generation model developed by Baidu, capable of text-to-image generation, referenced as a previous-generation model compared to newer alternatives.

An AI-powered image editing tool for relighting and brightness adjustment that uses generative AI to intelligently brighten dark images while preserving detail and avoiding artifacts, based on Flux2Cline architecture.

An open-source AI image upscaling tool released by Nvidia for enhancing image resolution with high accuracy.

OpenAI's image generation model that ranks as the top-performing model on the Arena AI image generation leaderboard, known for high-quality image synthesis and prompt adherence.

An image generation model that serves as the foundation for various image editing and generation tasks, designed to run on most consumer devices.

Google's AI image generation and editing model capable of creating and modifying images, integrated into the Gemini ecosystem.

A Google Workspace tool for editing existing photos, creating photos from scratch, and designing flyers and graphics, built on Google's AI image creation capabilities. It integrates AI-powered features such as cropping, object removal, and text overlay for graphic design.

An image generation model that creates images directly in pixel space rather than latent space, bypassing the traditional VAE conversion step used by most diffusion models.

An AI-powered design and image generation tool, now owned by Perplexity, that enables creative teams to quickly generate and iterate on visual designs.

An AI-powered creative design tool that allows users to generate and iterate on visual content quickly using generative AI capabilities.

A distilled, faster variant of Alibaba's Z-Image model optimized for quick image generation with high visual quality, particularly strong in realistic portrait photography.

The raw foundational model from Alibaba's Tong Yi Lab that serves as the base for the Z-Image family, capable of both image generation and image editing tasks.

A family of conversational voice models from OpenAI featuring full duplex interaction, allowing simultaneous speech processing and generation with variants including GPT Live 1 and GPT Live 1 Mini.

An AI vision model specifically designed to read and understand sheet music, capable of recognizing symbols, timing, pitch, staff lines, spacing, and structural elements. Trained on 9.7 million pages of sheet music covering around 400,000 musical works, it is lightweight at under 500 megabytes.

Google's conversational AI feature that enables extended, natural voice conversations, similar to a full-duplex voice assistant experience.

An open, modular speech-to-speech pipeline by Hugging Face that chains together speech recognition, language model inference, and text-to-speech into a real-time conversational AI system. It powers applications including the Reachy Mini robot.

A 2-billion parameter vision-language model from Alibaba's Qwen series that serves as a backbone for multimodal AI applications including robotics and visual understanding tasks.

A real-time conversational AI avatar system that handles text, audio, and video within a single transformer model, enabling natural two-way conversations with realistic facial expressions and gestures at 25 frames per second with approximately 200ms latency.

A lightweight AI model from Google, part of the Gemini family of multimodal AI models designed for on-device and efficient inference tasks.

An AI initiative by Planet Labs that indexes and analyzes Earth observation satellite data to make the planet searchable, analogous to how Google indexed the internet. It combines the world's largest Earth-observing satellite fleet with AI and data processing capabilities.

A multi-agent orchestration system by Sakana AI that dynamically routes tasks to the best-suited AI models from a pool of expert agents, balancing strong performance with low latency for everyday work.

AI-enabled smart glasses by Meta, co-branded with Ray-Ban and Oakley, featuring capabilities like Conversation focus which amplifies voices of people you're facing. Now requires a Meta One subscription for expanded access to premium features.

An iOS messaging app that uses on-device AI and augmented reality to let users send interactive AR characters through iMessage that can perceive and react to their real-world surroundings in real time.

A vision-language model by Allen AI that serves as the backbone for MolmoMotion, connecting language instructions to objects and points in images for multimodal reasoning tasks.

A world model AI startup that gathers data from the physical world and simulates it with accurate physics, producing rich interactive video from text prompts for use cases ranging from video-game creation to robotics.

Apple's rebuilt AI assistant integrated into iOS 27, featuring advanced on-device capabilities including expressive voices, upgraded dictation, Visual Intelligence, and writing tools powered by Apple Intelligence.

Apple's revamped AI assistant powered by Google Gemini models, featuring enhanced natural language conversation, personal context awareness, screen analysis, and a standalone chatbot-like app interface. It is deeply integrated across iOS 27, macOS 27, and watchOS 27, capable of creating reminders, writing and proofreading documents, generating shortcuts, and building custom Safari extensions using natural language.

A 4B-parameter multimodal safety classification model by NVIDIA that unifies image and text input evaluation, multilingual coverage across 140+ languages, custom enterprise policy enforcement, and auditable reasoning traces into a single inference call for content moderation.

An upcoming AI interface from Thinking Machines Lab designed to process continuous streams of audio, text, and video in 200-millisecond intervals, enabling real-time understanding of natural human communication patterns including interruptions and pauses.

A 3-billion parameter OCR model by Nanonets designed for document text extraction. It achieved notable degeneration rate reduction from 1.61% to 0.20% when trained with DPO techniques.

A family of computer-use AI agent models by H Company, available in multiple sizes (0.8B to 35B-A3B), designed to autonomously operate across web, desktop, and mobile GUI environments. It supports local and on-device inference with quantized checkpoints (FP8, Q4 GGUF, NVFP4) and integrates with various agent frameworks.

Apple's voice assistant receiving a major AI overhaul powered by Google's Gemini model, transforming it into a full AI chatbot with its own standalone app, on-screen context awareness, and the ability to perform actions across apps.

The 32B parameter variant of NVIDIA Cosmos 3, featuring a 32B reasoner and 32B generator designed for large-scale synthetic data generation and research, running on NVIDIA Hopper and Blackwell GPUs.

The 8B parameter variant of NVIDIA Cosmos 3, featuring an 8B reasoner and 8B generator optimized for efficient inference on workstation-grade compute such as the RTX PRO 6000 GPU.

An open-source vision-language grounding model by Nvidia that can detect and segment objects in images and videos using natural language queries, employing parallel box decoding for fast and accurate bounding box prediction. It has only 3 billion parameters and can run on most consumer GPUs.

An NVIDIA safety model that introduced reasoning-based content safety classification with step-by-step reasoning traces, serving as a precursor to the reasoning capabilities in Nemotron 3.5 Content Safety.

Apple's voice-activated AI assistant integrated across its devices, reportedly being enhanced with Google-powered AI capabilities ahead of WWDC 2026.

Microsoft's AI-powered assistant integrated across its products, offering conversational AI, content generation, and productivity features.

ByteDance's 3-billion-parameter open-source unified multimodal model capable of text-to-image, text-to-video generation, image and video editing, and visual understanding.

DeepSeek's AI research introducing visual thinking capabilities that allow AI systems to point at and reference parts of images during reasoning, reducing visual token usage by 90% while matching or beating frontier models on benchmarks. It uses policy distillation from expert models into a unified student model.

A family of transformer-based AI models by Allen Institute for AI (AllenAI) that processes satellite imagery for remote sensing tasks such as tracking mangrove change, classifying forest loss drivers, and producing crop-type maps. Version 1.1 cuts compute costs by up to 3x compared to v1 while maintaining similar performance.

Google's multimodal AI model featuring a large context window and strong performance across text, image, and video understanding tasks.

A lighter, faster variant of Google's Gemini Omni model, part of the new creation-focused model family that handles multimodal inputs and outputs.

AI interaction models developed by Thinking Machines Labs (founded by former OpenAI CTO Mira Murati) that feature real-time translation, intelligent interruption handling, visual monitoring, and simultaneous speech capabilities for natural human-AI conversation.

Google's AI assistant built on the Gemini Intelligence stack, integrated into Android devices and the Google Book, capable of handling complex tasks through natural interaction methods including mouse gestures and voice commands.

NVIDIA's open 30-billion-parameter multimodal AI model that processes images, video, and audio with exceptional throughput and cost efficiency, achieving nearly 10x real-time video processing speed — roughly 3x faster than QwQ 3 Omni and up to 7x faster on documents.

A suite of AI-powered features by Google built on the Gemini platform, designed for Android devices including Pixel and Samsung Galaxy phones, that enables multi-step task automation, contextual understanding from images, and intelligent agent capabilities across apps and services.

A full-duplex AI interaction system developed by Thinking Machines Lab (founded by former OpenAI CTO Mira Murati) that processes user input and generates responses simultaneously, enabling natural conversational AI with response times of approximately 0.40 seconds.

A full-duplex AI model from Thinking Machines Lab that can listen and respond simultaneously, achieving 0.40-second response times comparable to natural human conversation speed.

Alibaba's multimodal AI model capable of processing text, images, video, and audio, used as a benchmark comparison for throughput performance in multimodal tasks.

OpenAI's contrastive language-image pretraining model that learns to match images with text descriptions, commonly used as an image encoder in multimodal AI systems.

Amazon's rebooted AI-powered virtual assistant that can generate on-demand audio podcast episodes on any topic, featuring customizable AI host personalities and conversation styles. It sources content from over 200 news publications and delivers episodes to users' devices.

Amazon's enhanced AI assistant with improved intelligence and capabilities including smart home management and vacation planning, launched to all U.S. users.

A Tokyo-based startup founded by ex-Googlers that develops infrastructure to convert petabytes of unviewed video and audio into structured, queryable business data using vision-language models.

InfiniMind's flagship long-form video intelligence platform capable of processing 200 hours of footage to pinpoint specific scenes, speakers, or events, with beta release scheduled for March 2026.

An open-source real-time music generation tool released by Google that can generate music in real time.

ElevenLabs' AI-powered tool designed for marketing and branding teams to generate music and audio content for commercial use.

ElevenLabs' dedicated platform for creating AI-generated songs, allowing users to build tracks section by section including intro, verse, and chorus, then stitch them together.

Stability AI's latest family of audio generation models capable of creating professional-grade music compositions of over six minutes in length. The family includes four models ranging from 459M to 2.7B parameters, with smaller models suitable for on-device generation and larger models for full-length compositions.

An open-source AI voice cloning tool that can generate singing voices from just a few seconds of a reference voice, allowing users to make any voice sing any song with custom melodies and lyrics. It is lightweight (under 3GB) and can run on low-end GPUs or CPUs.

SambaNova Systems' next-generation AI inference chip unveiled in February 2026, designed to handle multi-trillion-parameter frontier models on a single rack with high-speed performance.

Meta's custom-designed AI chip program that develops modular accelerator chiplets for training and inference workloads, including ranking, recommendation algorithms, and broader AI applications across Meta's platforms.

A French AI research lab backed by telecom billionaire Xavier Niel, focused on open-source AI research including audio and voice models. It is the lab from which Gradium spun out.

A VPN service by AdGuard that follows a zero-logging policy and provides access to servers in more than 60 locations worldwide, supporting up to 10 simultaneous device connections for private browsing.

An ad-blocking and privacy tool by AdGuard that blocks intrusive ads, stops trackers, filters phishing and malware threats, and includes parental controls across major platforms including Windows, macOS, Android, iOS, and Chrome.

A planned AI data center infrastructure by SpaceXAI designed to operate in space, with demonstrations expected soon and commercial deployment targeted for 2028.

OpenAI's custom-built AI inference processor developed in partnership with Broadcom, designed to deliver better performance-per-watt efficiency compared to competitor chips.

SpaceX's satellite-based mobile connectivity service that aims to provide direct-to-device cellular coverage via Starlink satellites, positioning itself as a potential competitor to traditional wireless carriers.

An enhanced version of the Massive Multitask Language Understanding benchmark used to evaluate the knowledge and reasoning capabilities of large language models across a wide range of academic subjects.

A bipedal humanoid robot developed by Agility Robotics, designed for warehouse and logistics tasks. It is commercially deployed across multiple customer sites to address labor shortages and improve operational efficiency.

A human-validated benchmark by Allen AI designed to measure object-centric 3D motion forecasting accuracy, containing 2,700 video clips for evaluating motion prediction models.

Snap's latest augmented reality glasses featuring a proprietary liquid-crystal-on-silicon display, tetherless design without a computing puck, and AR computing capabilities including heads-up directions, translation, measurement tools, and shared computing experiences. Priced at $2,195.

A commission-free trading platform that enables retail investors to trade stocks, ETFs, and other securities. It reported record-breaking traffic during the SpaceX IPO debut.

A high-performance AI accelerator chip from NVIDIA used in data center GPU clusters for training and inference of large AI models.

A home assistance robot developed by Hello Robot, now in its fourth generation, designed to work in real homes with real people, featuring a telescoping arm, omnidirectional wheeled base, and voice-operated control for tasks like object manipulation and household assistance.

A $1.2 billion cloud computing and AI services contract shared between Google and Amazon to provide technology infrastructure to the Israeli government and military.

Nvidia's new PC CPU/superchip designed for AI agent PCs, offering 1 petaflop of AI performance with 6,144 Blackwell GPU cores and 20 CPU cores, capable of running large language models locally on Windows laptops.

A near-memory processing chip designed by startup XCENA that places compute capabilities closer to DRAM, aiming to reduce AI inference bottlenecks by handling data operations near memory without costly round trips between CPUs, GPUs, and memory.

Waymo's autonomous driving technology stack that powers its fleet of self-driving robotaxis. Now in its 6th generation, it enables vehicles to navigate various road and weather conditions, including snowy environments.

A free app by OverDrive that allows users to borrow ebooks and audiobooks from their local public library, compatible with Kindle and other e-reader platforms.

A book tracking app that helps readers log their reading progress, set goals, and discover new books based on mood and preferences. Launching a Kobo integration in June 2026 to automatically sync reading progress.

A benchmark created by Google DeepMind to comprehensively evaluate the ability of language models to generate factually accurate text, measuring both overall accuracy and search-augmented factual responses.

Microsoft's custom AI chip launched in January, designed to handle AI workloads in Azure data centers as an alternative to third-party GPU solutions.

Meta's $3.99/month paid subscription tier for Facebook that provides social expression and profile customization features similar to Instagram Plus.

Meta's $2.99/month paid subscription tier for WhatsApp that adds app themes, custom ringtones, premium stickers, and extra pinned chats.

Meta's upcoming broader subscription umbrella currently in early testing that will eventually house AI-focused tiers and professional plans for creators and businesses.

A tiny open-source robot made by Hugging Face, designed as a compact humanoid platform that can run various AI applications.

Nvidia's next-generation GPU platform that is sold both standalone and bundled with the Vera CPU for AI computing workloads.

Nvidia's CPU product purpose-built for agentic AI workloads, designed to process tokens as fast as possible rather than using traditional core-based cloud architecture, targeting a $200 billion market opportunity.

LetinAR's proprietary optical technology for smart glasses that arranges tiny optical elements inside a lens to direct light precisely into the user's eye, enabling brighter images in a thinner, lighter, and more power-efficient form factor than competing approaches.

A training dataset published alongside the Ettin Reranker models, consisting of a subset of lightonai/embeddings-pre-training mixed with a reranked subset of lightonai/embeddings-fine-tuning.

A pre-training dataset by LightOn AI used for training embedding and reranking models.

Google's app development platform providing backend services including Firestore database, Firebase Auth, and Firebase App Check, with planned integration into Google AI Studio's app creation workflow.

An online educational course created by Andrej Karpathy that teaches students how to build neural networks from scratch in code, covering fundamentals of deep learning and language models.

Sony's AI-driven robotic table tennis system that uses nine cameras and real-time spin tracking to play table tennis, becoming the first robot to defeat an elite human player.

Google's Android-based desktop operating system designed to power Googlebook laptops, built to integrate seamlessly with Android phones and optimized for Gemini AI.

A Google initiative targeting the mid-2030s to build data centers in space, leveraging orbital infrastructure for AI compute.

A quadruped robot by Mirami Technology designed to solve high-speed locomotion physics, serving as the research foundation for the Bolt humanoid robot.

A two-legged humanoid robot by Mirami Technology (Shanghai Robotics startup) that broke the world record for fastest humanoid robot at 10 meters per second (22.4 mph).

A benchmark measuring well-specified knowledge work tasks across 44 occupations, on which GPT 5.2 claims to be the first model at or above human expert level.

A benchmark measuring the ability of AI models to perform tasks in the terminal, particularly relevant for coders.

NVIDIA's family of open AI models and datasets designed for building AI agents, featuring over 10 trillion pre-training tokens and millions of post-training samples spanning multiple domains including code, math, safety, and agentic behavior.

A high-tier variant of Anthropic's Opus model line, representing one of the top-performing models in coding and reasoning benchmarks.

A compact vision-language-action policy built on Qwen3-VL-2B that uses a JEPA world model to predict future frames in latent space during training, providing world-model supervision at zero extra inference cost for robot control.

An autoregressive video-action model for robotics that predicts future video and actions together chunk by chunk, feeding real observations back in to keep its imagination grounded, runnable on a single 24–32 GB GPU.

A world action model for robotics that pairs a ~5B video-generation expert with a compact action expert in a single network, learning to dream rollouts during training but skipping imagination at inference to directly denoise action chunks.

NVIDIA's newest open cross-embodiment foundation model for robotics that uses Cosmos-Reason2-2B (built on Qwen3-VL) feeding a flow-matching action head for generalized robot control.

A vision-language-action model pretrained on interleaved vision-text-action data, featuring a Qwen2.5-VL-3B backbone with a flow-matching action head for robot policy learning.

A ~450M-parameter diffusion transformer policy for robotics based on Toyota Research Institute's Large Behavior Models recipe, conditioned on CLIP vision and language embeddings to learn multiple tasks selected via natural language.

An open-source frontier AI model developed by Chinese company Meituan, featuring 1.6 trillion parameters in a mixture-of-experts architecture with 48 billion active parameters. Notable for being trained entirely without NVIDIA GPUs, using AI ASIC superpods instead.

Microsoft's in-house AI models developed to power its own applications like Word and Excel, reportedly being used to reduce the company's reliance on external AI providers and cut costs.

A startup building its own open-weight AI models from scratch, part of the growing ecosystem of open source AI companies attracting venture capital investment.

A 32-billion parameter dense language model from the Qwen3 family developed by Alibaba, supporting tensor-parallel deployment across multiple GPUs.

A 235-billion parameter FP8 Mixture-of-Experts language model from the Qwen3 family developed by Alibaba, supporting data-parallel and expert-parallel deployment across multiple GPUs.

An AI benchmark and model used for evaluating coding and software engineering capabilities, considered one of the more accurate and realistic benchmarks in the field.

Anthropic's frontier large language model capable of coding, reasoning, and building complex applications, with support for multi-modal inputs and advanced problem-solving tasks.

A vision-language-action model included in the LeRobot v0.6.0 model zoo for robot policy learning.

An AI program developed by DeepMind researchers that became the first AI to defeat professional players at no-limit poker, using reinforcement learning techniques.

A transformer-based model developed by Allen AI that estimates both the density and score of a distribution in a single forward pass without retraining. It generalizes kernel density estimation and achieves significantly better accuracy in high dimensions for tasks relevant to generative modeling, Bayesian inference, and scientific computing.

Base44's proprietary large language model trained on tens of millions of real user interactions, designed to optimize latency, cost, and efficiency for vibe coding and app creation tasks.

A family of small AI models from Mistral AI optimized for edge devices such as smartphones and other resource-constrained hardware.

OpenAI's flagship and most powerful model in the GPT-5.6 lineup, featuring a 'max' reasoning effort mode and an 'ultra' mode that uses coordinated subagents to solve highly complex tasks. Priced at $5 per million input tokens and $30 per million output tokens.

OpenAI's mid-tier model in the GPT-5.6 family, designed as a balanced and efficient model for everyday use. It sits between Sol and Luna in terms of capability and cost.

OpenAI's smallest and fastest model in the GPT-5.6 family, optimized for fast and high-volume tasks. It is comparable to Anthropic's Haiku-class models in terms of sizing.

An enhanced reasoning mode for OpenAI's GPT-5.6 Sol that goes beyond single-agent capabilities by leveraging sub-agents to accelerate complex work, providing deeper reasoning on difficult tasks.

The premium variant of Sakana AI's Fugu orchestrator model, tuned for maximum answer quality on hard multi-step problems. It competes with top-tier models like Fable 5 and Mythos on key benchmarks.

A 7B parameter hybrid language model by Allen Institute for AI (AI2) that combines attention layers with recurrent layers, offering advantages on meaning-bearing tokens like nouns, verbs, and adjectives compared to pure transformer models.

A Japanese AI research lab focused on nature-inspired intelligence and recursive self-improvement, known for developing novel AI models and research into self-improving systems.

An AI model developed by an Asian AI company that approaches Mythos-level capabilities, representing growing competition in the frontier AI model space from non-U.S. developers.

A 550-billion-parameter hybrid language model by NVIDIA featuring Mamba2, LatentMoE, and Multi-Token Prediction architectures, designed for large-scale generative AI tasks requiring multi-node GPU training.

A 30-billion-parameter Mixture-of-Experts language model by NVIDIA with approximately 3 billion active parameters, designed for efficient single-node training and inference.

A specific variant of Google's Gemma 4 model family with 26 billion parameters using a mixture-of-experts architecture with 4 billion active parameters, designed for efficient local inference.

A 7B parameter transformer-based language model developed by Allen Institute for AI (AI2), designed for open research and benchmarking against hybrid architectures.

A 122-billion parameter mixture-of-experts language model from Alibaba's Qwen series featuring a hybrid Mamba/attention architecture with a 256K-token default context window.

A language-guided 3D motion forecasting model by Allen AI that predicts where object points will move in 3D space given a video frame, marked points, and written action instructions, built on the Molmo 2 backbone. It comes in autoregressive (MolmoMotion-AR) and flow-matching (MolmoMotion-FM) variants.

A flagship open-source large language model by Zhipu AI (GLM/ChatGLM team) designed for long-horizon tasks, featuring a stable 1M-token context window, advanced coding capabilities with flexible effort levels, and an improved sparse attention architecture with IndexShare. It is MIT-licensed and ranks as the highest-performing open-source model on long-horizon coding benchmarks.

Meta's video-based world model designed to understand, predict, and plan in the physical world by combining large-scale video data with robotic interaction data, aimed at building grounded AI understanding beyond language.

A New York-based AI startup building a foundation model that trains AI agents on spatial-temporal reasoning using a dataset of 2 billion gameplay videos per year, enabling agents to perceive, anticipate, and interact in simulated environments.

A large language model built on a fully subquadratic sparse attention architecture, offering a 12 million token context window with dramatically reduced compute costs. It processes tokens 52 times faster than flash attention and uses nearly 1,000 times less compute than traditional transformer-based LLMs, enabling reasoning over massive documents, codebases, and data collections.

A foundational AI model developed by Fin (formerly Intercom), described by its CEO as a groundbreaking model released as part of the company's AI-first transformation.

A policy checkpoint model that runs through the LeRobotLocal inference path, used for robot control and policy inference within the Strands Robots framework.

NVIDIA's generalist robot policy inference system that runs via Docker containers on GPUs, providing foundation model-based robot control capabilities requiring at least 16 GB of video memory.

An Indian AI startup focused on building indigenous AI models, contributing to India's efforts to develop homegrown AI technology.

An advanced AI model developed by Anthropic, reportedly capable of discovering critical software vulnerabilities in systems like Linux and Ghost. It was shut down by the US government due to national security concerns.

A 12-billion parameter dense language model by Google, part of the Gemma family of open models. It supports instruction-tuned variants and is compatible with standard transformers and PEFT workflows.

An open-source world simulation model released by NVIDIA that generates super realistic world simulations.

A frontier-class open-source AI model released by MiniMax, a Chinese AI company known for developing competitive large language and multimodal models.

Microsoft AI's flagship reasoning model developed in-house, designed for complex thinking and reasoning tasks. It competes with top-tier models like Anthropic's Claude Sonnet.

An AI system developed by Google DeepMind designed to solve complex mathematical problems by using formal proof languages like Lean, combined with a tournament-style iterative refinement process that uses judge AIs to progressively improve solutions.

A frontier AI model developed by Anthropic, evaluated using the Epoch Capabilities Index. Its system card found no evidence of a sustained AI-attributable 2× acceleration in AI progress.

NVIDIA's open-source AI model with 550 billion total parameters and up to 55 billion active per token, built on a hybrid Mamba transformer mixture of experts architecture. It is claimed to be five times faster and 30% cheaper than comparable frontier open models, with full open access to the model, training scripts, and data.

A frontier AI model developed by Anthropic, also known internally as Mythos. It is a large language model that was released with built-in silent sabotage mechanisms designed to degrade responses for certain queries related to AI model development, intended to prevent distillation attacks by foreign actors.

A 12B-parameter Mixture-of-Experts model by JetBrains, trained from scratch on natural language and code. It activates only 2.5B parameters per token, enabling efficient high-throughput, low-latency inference for tasks like routing, RAG, summarization, coding features, and private deployments, released under the Apache 2.0 license.

An advanced large language model by OpenAI in the GPT-5 series, released in late 2025 as part of the wave of models that enabled new agentic AI capabilities.

An advanced AI model developed by Anthropic, notable for its cyber capabilities. It is part of Anthropic's Claude model family and has been the subject of government safety testing discussions.

A large language model by Anthropic in the Claude Opus family, featuring a new 'dynamic workflow' tool that enables more flexible and adaptive AI task execution.

A Google open-weights language model available in GGUF format for efficient local inference, used as a recommended LLM for the Reachy Mini local voice pipeline.

Cursor's in-house AI coding model built on top of Moonshot's Kimi 2.5 base model, achieving near-frontier performance on coding benchmarks at significantly lower cost than competing models like Claude Opus 4.7 and GPT-5.5.

A compact 0.6-billion parameter language model from Alibaba's Qwen3 series, used to demonstrate delta weight sync where per-step payload drops from 1.2 GB to a fraction of that.

An open-source AI model project developed by Nous Research that serves as a foundation for building AI agents and assistants with advanced reasoning and function-calling capabilities.

Meta's flagship AI model that powers AI features across its family of apps including Facebook, Instagram, and WhatsApp.

Nvidia's GPU architecture designed for AI and high-performance computing workloads, adopted and deployed by every major hyperscaler, cloud provider, and model maker for data center AI infrastructure.

Proprietary physics-grounded AI models developed by SandboxAQ (an Alphabet spinout) that perform quantum chemistry calculations, molecular dynamics simulations, and microkinetics modeling for drug discovery and materials science applications.

A modernized BERT-based encoder architecture used as the backbone for building embedding models, offering improved performance for downstream tasks like retrieval and text representation.

A remote sensing foundation model that uses separate tokens per resolution when processing Sentinel-2 satellite data, demonstrating significantly better results with this approach compared to single-token methods.

A compact 97M-parameter multilingual embedding model by IBM that produces 384-dimensional embeddings and achieves the highest retrieval score (60.3 on MTEB Multilingual Retrieval) among open multilingual embedding models under 100M parameters. It supports 200+ languages, 32K-token context, and code retrieval across 9 programming languages under the Apache 2.0 license.

An open-weight large language model developed by Alibaba's Qwen team, capable of running locally on consumer hardware.

A family of on-device foundation models developed by Liquid AI, designed for efficient local inference on consumer hardware.

A small multilingual embedding model that serves as a baseline for sub-100M parameter multilingual retrieval, scoring 50.9 on MTEB Multilingual Retrieval across 18 languages.

Anthropic's new AI model release featuring agentic capabilities including 'agent swarms' and 'agent teams,' scoring nearly 30% on the APEX-Agents professional tasks benchmark in one-shot trials and 45% with multiple attempts.

A transformer architecture by François Fleuret at Meta that extends the classic decoder-based transformer with latent variables to make underlying decisions about sequence generation, such as generating consistent positive or negative movie reviews.

A Google Research architecture that learns to memorize at test time, enabling models to go beyond current context windows by maintaining memory across chunks of long sequences, presented at NeurIPS.

OpenAI's model described as representing an inflection point in AI capabilities. It ran uninterrupted for one week writing 3 million lines of code to create a browser from scratch, and solved multiple Erdős math problems.

Meta's most capable pre-trained base model to date, codenamed Avocado, developed by Meta Superintelligence Labs. It outperformed best open source base models and was competitive with leading post-trained models even before post-training.

A miniature model within the GPT-5.1 system that decides whether a user's query is worth spending extended thinking time on.

A compact English automatic speech recognition model optimized for browser-based inference via Transformers.js, serving as the default ASR model in the library's pipeline system.

Google's speech-to-voice translation AI feature that enables real-time language translation of spoken audio, available on select Pixel devices.

A frontier automatic speech recognition model by AssemblyAI designed for universal multilingual transcription, ranking among top performers on code-switched speech benchmarks.

A multilingual automatic speech recognition model by Deepgram designed for transcribing speech across multiple languages.

An ASR model by Mistral AI designed for speech-to-text transcription tasks.

A voice AI startup that builds small, low-latency models and its own orchestration layer to power automated customer support calls in Africa and the Middle East, handling localized dialects of English, French, and Arabic.

A dictation product by Google built into Gboard that enables voice-based typing across apps on mobile devices.

A voice-based typing and dictation application that turns spoken input into structured text for productivity workflows.

An AI-powered dictation application that converts speech to polished text, competing in the growing market of intelligent voice-to-text tools.

An AI-powered dictation app that converts speech to text, part of the growing ecosystem of intelligent voice transcription tools.

An AI dictation application that transcribes speech into text, competing in the voice-to-text app market alongside other AI-powered dictation tools.

OpenAI's latest advanced large language model (LLM), considered a frontier model released for wide public access and regarded as on par with Anthropic's Fable in capabilities.

A free, interactive AI-powered educational chatbot by Computer Biology Labs Inc. that simulates conversations with George Washington, trained on his speeches, correspondence, and Library of Congress archives. Users ask questions verbally and receive spoken and transcribed responses.

An open-source coding-focused AI model family available in multiple sizes (9B, 35B, and 397B parameters), with the largest version competing with frontier-level coding systems on benchmarks like SWE Bench Verified and Terminal Bench.

Mistral AI's consumer-facing chat and agent platform, comparable to ChatGPT, rebranded from its original name Le Chat.

OpenAI's flagship conversational AI model and platform, one of the most widely used large language models for text generation, coding, and general-purpose AI tasks.

Anthropic's primary platform for accessing and deploying Claude AI models, serving individual users, teams, and enterprise customers through various subscription tiers.

Advanced AI language models developed by Anthropic, referred to here under the codenames Fable 5 and Mythos 5. These frontier models were subject to U.S. export control restrictions that temporarily suspended access.

Anthropic's most powerful generative AI model tier available to consumers, offering significantly higher usage limits for Claude's capabilities including advanced code generation.

A DistilBERT model fine-tuned on the SST-2 sentiment analysis dataset, hosted on Hugging Face Hub, commonly used for text sentiment classification tasks.

Anthropic's AI assistant available as a desktop application, providing conversational AI capabilities for text generation, analysis, and coding assistance.

Anthropic's latest frontier AI model in the Mythos class, designed for complex, long-running tasks such as large-scale code migrations, writing, and agentic research. It features extensive safety guardrails and represents a significant capability improvement over previous Claude models like Opus 4.8.

An AI-powered search and answer engine API by Perplexity AI that provides real-time web search results with cited sources, designed for integration into applications.

A 30B-parameter Mixture-of-Experts coding model by Cohere with 3B active parameters, specifically designed for agentic software engineering tasks and complex code generation, released under the Apache 2.0 license.

OpenAI's premium $200/month subscription tier offering the highest usage limits and access to the most capable models, though reportedly losing money for OpenAI due to heavy user usage.

A fine-tuned AI model designed to help people with ADHD overcome task-initiation paralysis by generating warm, dopamine-sparking conversational nudges instead of traditional to-do lists. Built on Gemma 3 12B with LoRA fine-tuning, it reads conversational context and provides one small actionable step to help users get unstuck.

A super efficient AI model from Google designed to run offline on consumer laptops without requiring cloud connectivity.

An enterprise AI platform that provides generative AI tools for businesses, including large language models and research into AI memory systems and personalization.

A 24B dense parameter coding model by Mistral AI designed for software engineering tasks.

A code-focused AI model from Microsoft's MAI family designed for code generation tasks, available in Microsoft Copilot and VS Code.

An open-source large language model available through various AI platforms, part of the Qwen model family developed by Alibaba.

A 4-billion parameter small language model developed by OpenBMB, designed for efficient on-device and resource-constrained inference tasks.

A 4-billion parameter small language model developed by NVIDIA, optimized for efficient deployment in agentic and conversational AI applications.

A 0.5-billion parameter small language model from Alibaba's Qwen series, suitable for fine-tuning and lightweight deployment in resource-constrained environments.