AI Tools Directory

A world model platform developed by Fei-Fei Li's World Labs that creates explorable 3D environments from visual inputs, with applications in video games, CGI effects, and media creation.

World Labs' product for creating explorable 3D environments from media inputs, with applications in video games, CGI effects, and spatial intelligence demonstrations.

Nvidia's simulation and digital twin platform used for building, operating, and testing physically accurate virtual environments, particularly for robotics and industrial applications.

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 benchmark leaderboard for evaluating and ranking automatic speech recognition (ASR) models on standardized tasks and datasets.

A leaderboard maintained by the ELM Research Center that ranks Arabic automatic speech recognition systems by average word error rate across six standardized test sets.

An enterprise app marketplace by OpenAI where partners including Adobe, Figma, Salesforce, HubSpot, and others can market their AI-integrated services, with customers able to apply their OpenAI commitment toward approved partner software.

A cloud-based platform that aggregates 50+ AI models into one workspace, allowing users to compare side-by-side responses from up to six models simultaneously, generate AI images, and access models from providers like OpenAI, Anthropic, Google, and others.

A searchable directory of user-shared examples and practical guides for AI agents, including task ideas, sample prompts, and links back to original posts demonstrating real-world use cases.

A comprehensive time series forecasting benchmark and leaderboard designed to evaluate foundation models across diverse forecasting scenarios, measuring metrics like CRPS and MASE for zero-shot and pretrained models.

A platform that makes open AI research accessible by linking papers to code, datasets, and state-of-the-art benchmarks, enabling discovery of research artifacts across AI domains.

A Chinese open-source AI model hosting platform that serves as a repository for AI models, functioning as a competitor to Hugging Face for sharing and accessing machine learning models.

A leaderboard hosted on Hugging Face for evaluating automatic speech recognition models, featuring held-out test sets designed to measure real-world performance beyond standard benchmarks.

A large-scale multilingual speech corpus derived from European Parliament recordings, widely used as a benchmark dataset for automatic speech recognition research.

A next-generation AI wearable device by Amazon's Bee that can be worn on the wrist or pinned to clothing, featuring dual microphones, IP67 water resistance, seven-day battery life, and AI-powered conversation summarization capabilities, priced at $74.99.

A credit-card-sized AI device by Amazon's Bee that magnetically attaches to the back of a smartphone, providing AI-powered conversation capture and summarization at a $99.99 price point.

A wildlife-deterrent AI startup spun out of KTH Royal Institute of Technology that uses edge devices with computer vision and on-device models to identify animal species and emit species-appropriate sounds to keep wildlife away from trains, airports, roads, and other infrastructure.

An AI-powered feature by Meta that allows Facebook and Instagram users to influence their content recommendations by describing their preferences in natural language, with changes reflected in real time.

An AI-powered platform that tracks how stories and rumors spread across more than 500 mainstream and fringe websites and social media platforms, helping organizations detect and prevent social media crises by analyzing behavior, relationships, and information pathways rather than just mentions and impressions.

An Australian AI startup that deploys thousands of AI agents disguised as potential victims to engage scammers across calls, messages, emails, and social media, gathering fraud intelligence such as bank accounts, crypto wallets, and scam tactics to help banks, telecoms, and government agencies prevent fraud.

A private workplace messaging app with built-in AI features including AI-powered search, voice chat, voice note and call transcriptions, and AI agents for workflow automation, designed to replace consumer chat apps like WhatsApp in business settings.

An autonomous customer knowledge platform that uses AI agents to continuously research, collect, organize, and update B2B sales prospect data from over 100 third-party sources, automating prospecting and CRM enrichment for revenue teams.

A proprietary AI platform developed by Scimplify that coordinates chemical manufacturing processes by identifying suitable production routes and facilities, while tracking quality and regulatory requirements using specialized AI agents for research, manufacturing, and quality control.

A Markdown-based editor feature within Figma that lets users define their design language in a structured document, serving as a reference for both human teams and AI agents to adhere to a brand's design rules while creating designs.

An open-source headless content management system acquired by Figma, now integrated with Figma Make to enable dynamic website generation and direct content management.

Google DeepMind's AI weather forecasting model that offers advanced capabilities including cyclone forecasting, increasingly used by the U.S. government for hurricane season response.

An AI-powered communications and PR platform that automates tasks like writing announcements, finding journalists to pitch, identifying speaking opportunities, and building media lists, designed to bring senior-level PR expertise to anyone.

An AI startup that generates code-compliant and permit-ready architectural construction blueprints in minutes from 2D sketches or PDFs, covering architecture, structural, mechanical, electrical, plumbing, and interior design.

A smart email cleanup tool by MacPaw integrated into CleanMyMac that uses metadata analysis to bulk-purge Gmail clutter, clear heavy attachments, and unsubscribe from junk mail without reading private message contents.

A desktop file conversion application that supports over 2,900 conversion types across images, video, audio, documents, and more, with built-in local AI tools for audio transcription, photo upscaling, and background removal — all processed on-device without cloud uploads.

An AI-powered creative tool included with Amazon Kids+ that allows children to use their voices to create interactive stories on Amazon's new Kids Tablets.

A compact AI-powered recording device by Plaud designed for portable conversation capture and AI-driven summarization, priced at $129.

An AI-powered coding assistant platform that helps developers write, review, and debug code, offering tools for code generation and search.

An AI company focused on building foundation models for understanding the physical world, enabling applications in physical AI and real-world sensing.

A cloud infrastructure provider offering GPU-powered cloud computing, storage, and AI/ML infrastructure services for developers and enterprises.

An educational platform founded by Dr. Marily Nika that offers training and certification programs focused on building AI products, including courses on AI agents and product management.

A biosecurity nonprofit organization that conducts narrow safety reviews and evaluations of AI systems, focusing on biological risk assessment.

JPMorgan Chase's AI-powered tool designed to forecast cash flow and automatically move customer funds between accounts to maximize yield, with plans to anticipate bills and budgeting needs.

An AI interviewer developed by HackerRank for technical hiring that can observe candidates completing technical assignments, ask follow-up questions, and evaluate problem-solving approaches.

An AI inference platform that provides fast and scalable access to generative AI models, used by developers and creators for product-building applications.

An AI cloud infrastructure startup that provides GPU cloud computing services for training and deploying AI models, reaching $120M in annual recurring revenue.

Microsoft's security solution for AI agents that provides agent identity management, observability, risk containment, and enterprise-level security for autonomous AI workflows.

An autonomous mode within Microsoft Copilot that can independently handle file management and email workflows without continuous user input.

A bird detection and deterrent system by Robin Radar that uses species-specific distress calls to disperse birds from airports and other infrastructure, while recording the activity.

An acoustic gunshot detection system that uses a network of microphones (approximately 25,000 as of 2024) to identify and locate gunfire in urban areas, alerting law enforcement in real time. Now operated under SoundThinking.

An AI evaluation and testing platform that helps enterprises assess, monitor, and improve the reliability and safety of their AI models and applications.

A data processing and AI infrastructure platform that enables real-time data pipelines and AI applications, helping companies build live AI systems over streaming and changing data.

A cloud-based physical security platform that uses AI-powered cameras, sensors, and access control systems to provide intelligent building security and monitoring.

A lightweight, affordable mobile manipulator robot built by Hello Robot, designed for real-world home and research applications including assistive tasks and manipulation.

A Hugging Face tool integrated with HuggingChat that automates machine learning workflows including dataset labeling, model training, evaluation, and publishing. It plans work, manages compute budgets, and runs on Hugging Face hardware.

A personal AI assistant startup founded by Zach Yadegari that combines a software agent with a $179 wearable band, capable of ordering food, booking travel, answering questions, and managing tasks via iMessage and a forthcoming app.

A Mac app by Google that serves as an AI-powered meeting note-taker, working completely offline using on-device models to capture, transcribe, and summarize meeting notes across apps including in-person meetings.

An AI-powered document processing platform that can parse, extract, and split complex documents with high accuracy for enterprise use cases.

A $99 smart ring by AI and hardware startup Natura that lets users interact with AI agents via voice commands with a finger press, while also functioning as a health tracker monitoring heart rate, sleep, and activity.

A Mac security app by MacPaw that combines a malware scanner, real-time protection, and an AI-powered Scam Detector to catch threats that bypass Apple's built-in defenses. It is independently certified by AV-TEST and offers on-demand scanning, system protection reviews, and scam analysis for suspicious texts, emails, and job offers.

An iterative generate-evaluate-refine inference strategy developed by NVIDIA that improves AI model outputs through test-time compute, enabling models to achieve significantly higher scores on competitive programming benchmarks.

An AI platform that reads patient records and uses language models to identify hospitalized patients at risk of undiagnosed conditions such as malnutrition, delirium, and aspiration pneumonia. It integrates with electronic health record systems to analyze clinical notes, lab results, and vital signs, generating daily risk scores for care teams.

An AI-powered game-creation platform from Google Labs that allows users to build browser-based 2D and 3D games using simple text prompts without coding expertise. It supports single-player and multiplayer experiences and includes features like leaderboards and a public Explore gallery.

An upcoming integration for Google Playground that promises professional-level game mechanics, high-fidelity 3D capabilities, and the flexibility of the Unity runtime for AI-assisted game creation.

A feature in Microsoft's revamped Windows 11 that enables sandboxing of AI agents, making it easier and safer to run and orchestrate AI agents locally on Windows PCs.

An AI-powered ambient listening app founded by former Plaud executive Peter Mo that provides real-time intelligence, insights, and proactive suggestions by continuously capturing audio context across iPhone, Mac, and Apple Watch devices.

A web traffic monitoring and analytics tool by Cloudflare that tracks internet traffic patterns, including automated bot traffic, across networks worldwide.

A suite of free, open-source creative tools built with AI assistance in Rust by developer Brandon Thomas, designed to replicate Adobe Creative Suite applications including Photoshop, Premiere, and After Effects alternatives.

An AI engine that predicts consumer behavior and the motivations behind it by building a foundation 'world model' that simulates why humans do what they do and how behavior changes over time, rather than relying on fine-tuned LLMs.

Anthropic's program offering startups subsidized access to Claude AI models, including a free year of Claude Team, $1,000 in API credits, access to Claude Marketplace, and virtual office hours with Anthropic's Applied AI team.

Anthropic's platform that allows developers and companies to build and distribute plug-ins for the Claude AI service.

An open source document editor by The Document Foundation that has deliberately chosen not to include any AI or generative AI features in its default installation, positioning 'no AI' as a privacy and data control feature. Users can optionally add AI via extensions connecting to local models.

An AI platform for generating ad campaigns, images, and videos, described as an 'agents lab for creative work' that enables users to create marketing assets using plain language. Founded by ex-Ramp engineers.

An AI personal assistant by Hark that uses a model trained specifically for computer use, allowing it to navigate the web and perform digital tasks on behalf of users across email, calendar, and other services, with a focus on privacy and transparency.

A robot vacuum by Shark that uses AI-powered navigation and object detection for autonomous floor cleaning, featuring a self-emptying base and up to 120 minutes of runtime.

A startup founded by Dr. Ding Zhao (Carnegie Mellon University), Kyle Wong, and Simo Rachidi that evaluates the safety of generative AI-powered robotic systems by running thousands of simulated scenarios with realistic human models in digital environments.

An AI-powered advice app for young women that provides personalized relationship and dating guidance grounded in behavioral science, designed to feel like chatting with a trusted friend while incorporating safety safeguards.

HackerRank's AI interviewer agent that conducts technical job interviews, observes candidates as they work through real-world coding tasks, and evaluates critical thinking, judgment, and AI fluency in addition to correctness.

A conversational AI agent built into TikTok that helps users discover and purchase products by providing real-time shopping guidance, including product details, shipping information, sizing, availability, and purchase completion assistance.

A browser-based screen recording tool that offers instant shareable links, automatic transcripts of recordings, clickable CTAs with analytics, and annotation features. It works as a Chrome extension on Windows, macOS, Linux, and ChromeOS.

An AI safety organization focused on evaluating and ensuring the safe development and deployment of AI systems.

A benchmark by Microsoft that evaluates AI agents by grading the terminal backend state and side effects they leave behind across 507 stateful business workflows, running each task 20 times to measure reliability and consistency.

An AI assistant that lives in iMessage and RCS, connecting to users' calendars and conversations to identify actionable items, add events, set reminders, and perform research tasks.

An AI assistant accessible through iMessage, WhatsApp, and Telegram that runs on its own private cloud computer, capable of managing reminders, researching topics, tracking flights, handling email, and executing multi-step tasks.

Google Cloud's fully managed, serverless data warehouse and analytics platform that supports AI and machine learning workloads at scale.

An AI safety testing platform that uses simulated AI agents mimicking diverse user demographics to red-team AI models for psychologically harmful interactions, providing auditable safety scores for high-risk AI applications.

An open-source project by Meta that provides firmware and a Linux SDK allowing developers and hobbyists to build custom hardware devices that connect to Meta's Muse AI agent.

A USB-C-powered hardware device built by Meta that connects the Muse AI agent to a home network, enabling it to communicate with smart devices including speakers and smart TVs.

A benchmarking index (versions 0.2.1 and 0.3) used to evaluate and compare the performance of AI decision models across tasks like reading comprehension, toxicity detection, intent classification, medical QA, and cross-lingual understanding.

A pipeline developed by ServiceNow CoreAI that automatically generates training data for enterprise AI agents by identifying capability gaps in a target model and creating validated tasks to address those weaknesses.

An open-source training infrastructure framework by the Allen Institute for AI (AI2) designed to scale mixture-of-experts (MoE) language model training into the trillion-parameter range while maintaining computational efficiency.

A Shopify site-building tool that lets merchants create and customize their online stores by chatting with an AI agent, rendering real code in real time rather than static previews.

Shopify's AI assistant that helps merchants build apps, customize themes, create sites, and make edits to their online stores through a conversational chat interface.

Meta's consumer AI agent app that has reached the No. 1 position on app stores, designed to carry out tasks on behalf of users.

A developer platform that provides tools including a unified API, extensible channel framework, CLI, and observability suite for building AI agents that operate over messaging services like iMessage, WhatsApp, Telegram, SMS, and email.

A software platform that enables AI models to run directly on satellites, allowing spacecraft to process data onboard rather than downlinking to Earth, reducing transmission costs and enabling distributed cloud computing across satellite constellations.

An open-source decision model by Amazon that sorts between pre-decided options and delivers calibrated confidence scores, designed for use in agentic workflows as a fast, low-cost alternative to full LLMs.

A PDF editing application by Readdle for Mac that includes OCR text recognition for scanned documents, along with editing, annotation, conversion, and form-filling capabilities.

An AI-powered audio detection system by Flock Safety (formerly known as Raven) that continuously analyzes ambient audio to identify specific sound events such as gunshots, fireworks, street takeovers, and vehicular accidents, sending real-time alerts to relevant parties.

An AI startup that models human behavior, notable for raising a massive $480 million seed round at a $4.48 billion valuation.

A privacy-focused AI assistant positioned as a competitor to Instinct and Muse, launched by a young Silicon Valley founder and designed to prioritize user privacy.

An AI startup that helps ad agencies and brands generate creative assets and campaigns using artificial intelligence.

A cloud-based data platform that combines data storage, processing, and analytics into a single service, running on major public clouds like AWS, Azure, and Google Cloud. Businesses use it to store large amounts of structured and unstructured data, build data pipelines, train machine learning models, and share data with partners.

A cloud AI compute platform by Cerebras Systems that provides inference and training services powered by the company's wafer-scale processors, designed for demanding AI workloads.

The latest generation of Cerebras Systems' wafer-scale AI infrastructure, introduced in August 2026, designed to deliver high-performance compute for AI training and inference workloads.

A Berlin-based durable workflow infrastructure platform that provides a resilient execution engine for multistep workflows, making them crash- and interruption-proof — particularly well-suited for AI agent processes.

An AI-powered search feature within the Airbnb app that allows users to search for properties via text or voice prompts, generating dynamic filters based on preferences and providing AI-powered summaries and review comparisons.

An API by OpenAI that provides a super-powered classifier built on an LLM, allowing developers to give a model a predefined set of options to choose between—such as image classification categories or agent behaviors—at high speed and low cost.

An AI operating system by Destro AI built on open-weight vision-language-action models that interprets camera images and instructions to control robot movements in logistics and warehouse environments.

A San Francisco-based startup that offers AI agents for hardware design, automatically aligning CAD drawings with product requirements, simulation results, and testing data.

An AI-powered marketing platform that helps advertisers explore geo-based experiments and optimize their advertising strategies.

A $3,499 personal AI computer designed for individual use, created by a startup that raised $11M in funding from a 19-year-old founder.

An AI-powered platform for retrosynthetic analysis that helps chemists plan the synthesis of molecules, founded by Piotr Byrski.

A domain-scanning service that allows researchers to monitor web traffic and track AI agent activity by recording queries made by agents that use it to load websites they cannot access directly.

OpenAI's personal AI agents that can be used individually or in collaborative workspaces where multiple Dots and humans work together on tasks.

OpenAI's text watermarking method that subtly shapes a language model's word choices to embed an invisible, machine-detectable pattern in AI-generated text, developed in collaboration with researchers from the University of Pennsylvania and Yale.

OpenAI's system for embedding invisible statistical watermarks into AI-generated text from ChatGPT and Codex, enabling machine-readable detection of AI-authored content in compliance with the EU AI Act.

A post-generation verification layer for MCP-based LLM agents that checks whether claims in an AI-generated answer are attributed to the correct source, rather than merely verifying factual accuracy against pooled evidence. It preserves source identity through decomposition, routing, support scoring, and attribution checking to detect cross-source conflation errors.

A personal AI assistant founded by former Yahoo CEO Marissa Mayer that builds context about users by analyzing their camera roll photos to provide personalized recommendations for activities, vacations, gifts, and more.

An AI security platform that uses context graphs to discover, map, and secure AI agents, SaaS applications, and their interconnections across enterprise environments, helping security teams manage permissions and access controls.

A set of security tools by OpenAI that allows developers to scan GitHub repositories on demand or on a schedule, investigate findings, remove duplicates, and prepare fixes automatically in the cloud.

An OpenAI API for real-time decision-making that uses the Luna model to answer user-defined questions with predefined answers.

OpenAI's API for building AI agents, updated to support computer use and integration with Amazon's Bedrock Managed Agents to run OpenAI agents entirely on AWS.

OpenAI's developer tool for building ChatGPT plug-in extensions, allowing developers to create and submit interactive app-like experiences within the ChatGPT ecosystem.

A Boston-based voice intelligence platform that uses over 100 small AI models to provide enterprises with transcription, emotional analysis, deepfake detection, AI music detection, and policy enforcement for voice agents in regulated industries. It specializes in analyzing the intent and nuance behind human conversations and detecting synthetic voices.

An AI-powered platform that helps insurance agencies and brokers automate administrative paperwork for commercial insurance, including evaluating and recommending policies across over 250 coverage types. It serves over 300 insurance agencies, including 25% of the top 100.

A protocol by Shopify that enables browser-based AI agents to interact with Shopify storefronts and checkout processes, allowing agents to search products, manage carts, and complete purchases on behalf of users through structured APIs rather than HTML scraping.

A toolkit from Nvidia combining software and hardware products that adds independent security layers around AI agents to ensure they remain within their designated environments and cannot break out of testing sandboxes.

Nvidia's open source software that controls what AI agents can access while they operate, providing a software boundary to contain agent behavior within defined parameters.

Nvidia's independent hardware-level monitoring system that runs on BlueField-4 data processing units, continuously monitoring AI agent behavior and quarantining agents that attempt to move outside their boundaries.

Meta's enterprise AI platform that brings its full technology stack—including Muse, Meta Business Agent, Muse API, and Muse Code—to businesses and developers for deploying AI-powered products and services.

A San Francisco-based startup that builds compact on-device AI models for real-time deepfake voice detection, designed to run directly on smartphones without sending audio to the cloud.

An inference infrastructure startup specializing in serving AI models for video and image generation workloads.

OpenAI's always-on AI agent platform powered by GPT-6 Astra that runs on its own private cloud computer, learns from user feedback over time, and connects to over 4,000 apps via plugins to handle tasks 24/7 on behalf of users.

An AI-powered health scanning company that uses full-body scans to detect health issues, having raised $700 million in funding.

An environment developed by ServiceNow for training and evaluating enterprise AI agents, used to illustrate and test the AutoSynthData pipeline for generating agentic training tasks.

A software development kit by Vercel for building chat-based AI interfaces and applications.

An AI agent product by Tencent that uses Photon as its underlying messaging communication layer.

A social introductions app that uses AI agents to facilitate connections between users, built on Photon's messaging platform.

An AI-powered finance assistant that operates through messaging platforms, built using Photon's agent infrastructure.

An AI-powered email client that leverages agent-based interactions for managing email communications.

Apple's on-device AI-powered audio processing tools included in newer Apple Watch models, providing intelligent audio features powered by the S11 chip.

An event networking app by TechCrunch that uses AI-powered matchmaking to help attendees discover relevant people and schedule 1:1 meetings based on their interests and goals.

TCL's AI-powered image processing engine built into its TVs that automatically adjusts picture clarity, contrast, color, motion, upscaling, and backlighting based on the content being viewed.

An AI-powered sports tracking camera that physically moves to follow the action on the field in real-time, recording games in 4K at 30fps or 1080p at 60fps without requiring a subscription service or post-processing.

A durable execution platform founded in 2019 that enables developers to build reliable, fault-tolerant distributed applications and workflows, valued at $12.55 billion.

A text-to-speech evaluation benchmark that provides standardized test sets with reference audio for English and Chinese, used to assess TTS model quality including voice cloning capabilities.

Reddit's developer platform that allows third-party developers to build apps, bots, and integrations for Reddit communities, serving as the replacement for Reddit's public API.

A personal AI agent by OpenAI designed for both consumer and enterprise use, with applications for software engineers and agency creatives.

An always-on AI agent introduced by OpenAI at Dev Day 2026, designed to assist users with ongoing tasks and interactions.

An AI-powered government website built by DOGE alumni that runs on xAI's Grok, serving as a digital interface for U.S. government information.

Shield AI's platform-agnostic mission autonomy software designed for autonomous aircraft and drone operations, selected as an autonomy provider for the U.S. Air Force's Collaborative Combat Aircraft drone prototype program.

Waabi's autonomous driving technology designed for self-driving trucks and robotaxis, developed using simulation-first approaches and partnered with Uber for deployment of 25,000+ robotaxis.

Waabi's high-fidelity simulation platform used to train, test, and stress-test autonomous driving systems in a virtual environment before real-world deployment.

A team messaging platform designed for both human and AI agent collaboration, giving AI agents their own identities, inboxes, and the ability to participate in conversations, channels, DMs, and live calls alongside human workers.

A smart birding platform that uses AI-powered species recognition to identify over 99% of North American backyard bird species. It connects to motion-activated camera feeders, baths, and birdhouses, and now offers a Roku TV app for livestreaming and browsing clips sorted by species.

A language-learning platform that uses AI and gamification to help users learn new languages through interactive lessons, featuring its iconic owl mascot Duo.

A fine-grained factuality checker designed to verify whether claims made by language models are supported by provided evidence.

A factual consistency evaluation tool that assesses whether generated text is supported by reference information.

A factual consistency checker for summarization that evaluates whether generated summaries are faithful to their source documents.

A U.S. government AI-powered portal operated by the General Services Administration that serves as a single front door to the federal government, searching across roughly 29,000 government websites to answer citizen queries using Google Gemini and Grok AI models.

A framework for retrofitting language model outputs with attribution by retrieving supporting evidence and revising claims to be better grounded in sources.

An AI security startup that helps organizations discover files and data that have become accessible to AI tools and agents, identifying potential security risks in enterprise environments.

OpenAI's integration feature that connects ChatGPT to everyday productivity tools like Slack, SharePoint, Airtable, and Google Drive, enabling the AI to access and work with external data sources.

An AI agent that contributed to the rapid progression of agentic AI in 2026, part of the wave of personal AI assistants designed to handle tasks autonomously on behalf of users.

A Python framework by NVIDIA for writing high-performance, GPU-accelerated kernels. It enables developers to author statically typed kernels in Python that compile for CPU or CUDA execution, with support for differentiable kernels, DLPack interop with PyTorch/JAX, and deterministic execution modes.

An AI-powered feature within the YouTube Music app that allows users to create personalized stations, find music based on mood, discover songs that inspired specific tracks, find related podcasts, and get detailed information about songs and artists through natural language queries.

An AI-generated personalized audio guide within YouTube Music that provides weekly spoken previews of recommended podcast shows, explaining why they might appeal to the listener and linking directly to suggested episodes.

YouTube's creator management app that includes AI-powered features such as Ask Studio (an AI Q&A feature), AI-generated thumbnails and titles, draft feedback with suggestions on pacing and storytelling, dynamic thumbnail testing, and enhanced analytics with AI-driven insights.

An AI-powered assistant within YouTube Studio that can parse a creator's back catalog of videos and help optimize previous content with new thumbnails and A/B testing features, now available on mobile.

An enterprise AI platform that deploys coordinated teams of AI agents (called 'AI employees') to automate multi-step business processes across HR, IT, and finance, integrating with a company's existing applications and drawing on over 150 AI models.

A palm-sized Tamagotchi-like wearable device by Meta that houses the Muse AI agent, featuring a tiny screen displaying a digital avatar, a fingerprint sensor for instant activation, and real-time voice interaction capabilities without needing to unlock a phone.

A biotech startup that uses AI to discover new drugs derived from plants and microbes in the natural world, currently advancing AI-discovered drug candidates into human clinical trials for conditions including severe skin diseases and weight loss maintenance.

An Apple camera feature that uses machine learning to automatically detect and keep users centered in the frame during video calls, even as they move around.

A companion app by Therabody that pairs with its recovery devices and wearables, using an AI-driven Coach feature to analyze real-time health data and automatically build customized, step-by-step muscle recovery routines.

An AI-powered editing assistant coming to YouTube Shorts and YouTube Create that helps creators edit footage into videos using natural-language conversation, capable of selecting best takes, adding transitions, text, and more.

A YouTube feature that uses algorithmic personalization to show different thumbnail variations to different viewers based on what YouTube's algorithm predicts will most resonate with each individual user.

Meta's developer platform that enables third-party developers to build integrations and connectors for the Muse AI agent, allowing it to interact with external apps and services.

A feature by Nikon that allows camera users to download and apply settings presets created by professional photographers, enabling quick optimization of camera configurations for specific shooting scenarios.

H Company's API platform for accessing and deploying their agentic AI models, including the Holo4 series. It provides pricing and inference capabilities for running agentic workflows.

An academic benchmark for evaluating AI models on desktop control tasks and long agentic workflows. It features a public leaderboard tracking the performance of various open and closed models on computer-use scenarios.

A startup building an AI operating layer for insurance brokerages, helping brokers automate and streamline their workflows in the insurance industry.

A startup focused on providing AI-powered tools for insurance brokers, competing in the space of automating insurance brokerage operations and workflows.

Meta's API offering that allows developers and businesses to integrate Muse AI assistant capabilities into their own applications and workflows.

An AI-powered voice security and authentication platform that specializes in detecting deepfake voices and verifying caller identity for enterprises.

An AI voice platform that offers voice cloning, speech synthesis, and deepfake voice detection capabilities for enterprises and developers.

A Microsoft Azure cloud service that provides AI-powered content moderation and safety tools, including detection of AI-generated and manipulated content such as deepfake voices.

A Microsoft-owned AI company specializing in conversational AI, speech recognition, and voice biometrics, used in healthcare, enterprise, and security applications.

A cloud-native managed lakehouse platform that simplifies data lake management and provides infrastructure for analytics and AI/ML workloads.

A large-scale visual database created by Fei-Fei Li and collaborators, designed for use in visual object recognition research and widely credited with catalyzing the deep learning revolution in computer vision.

A structured format developed by EvalEval that combines benchmark metadata, evaluation-run data, and model metadata into interpretable records for understanding AI model evaluation results.

An open-source local AI tool for running large language models on personal devices, powered by the llama.cpp inference engine.

An open-source tokenization library by Hugging Face that converts text into token IDs for machine learning models, supporting BPE, WordPiece, and Unigram tokenizer families. Version 1 introduces major performance improvements including SIMD-based splitting, thread-local caching, and zero-allocation merge loops.

An AI-powered real-time bookkeeping platform for small businesses that uses Plaid to import live account data and automatically generates financial dashboards, profit-and-loss statements, and handles paperwork without traditional accounting software.

A natural language interface for the Tabby bookkeeping platform that allows users to interact with their financial data through conversational queries.

An AI startup focused on healthcare that uses artificial intelligence to assist with medical documentation and clinical conversations.

A Palo Alto-based edtech startup that transforms textbooks and PDFs into TikTok-style scrollable feeds using AI-generated video, audio, text, and interactive quizzes to make educational content more engaging for younger learners.

A PDF editing and conversion tool offering 20+ features including text editing, format conversion, on-device OCR for scanned documents, digital signatures, and AES-256 encryption, all processed offline without cloud uploads.

A browser-based video editing platform that combines four editing tools in one interface, featuring AI-powered highlight detection, automated captions, AI voiceovers, smart cropping, and 4K output for gaming clips, streams, podcasts, and other long-form video content.

A web service that acts as a browser proxy for security research, allowing users to analyze URLs without directly opening them, while publishing public logs of the activity.

An AI-powered automatic pet feeder by Petlibro that uses a built-in camera with AI recognition to identify up to 10 individual cats, track their eating habits, and provide personalized feeding data through a companion app.

An AI benchmarking platform that evaluates AI models on their ability to complete complex, industry-specific tasks across domains like law, finance, coding, cybersecurity, and mental health. Unlike traditional benchmarks, Vals keeps its test materials private to prevent companies from training against them, and measures real-world task performance rather than abstract intelligence.

A learning app that condenses nonfiction and fiction bestsellers into 15-minute text and audio summaries, offering personalized self-growth plans, microlearning sessions, and gamified progress tracking with over 2,000 book summaries available.

A Google Gemini feature that allows the AI to call businesses on users' behalf to inquire about pricing, services, and availability.

A Google Pixel phone feature that uses AI to navigate phone trees and wait on hold until a real person comes on the line, then alerts the user when a live representative is available.

A Google Pixel phone feature that uses AI to navigate automated phone systems and connect users directly to a live customer service representative.

A Google Pixel phone feature that displays on-screen options for navigating IVR (Interactive Voice Response) phone menu systems, making it easier to reach the right department.

A popular health and fitness app for tracking nutrition, exercise, and calories that acquired Cal AI in December 2025 to enhance its AI-powered food tracking capabilities.

An AI infrastructure platform that simplifies access to GPU infrastructure, enabling developers to deploy AI software across public cloud, private cloud, and on-premises environments without vendor lock-in. Acquired by Nvidia in July 2024.

Robinhood's AI-powered trading feature that allows users to trade equities, options, and crypto through AI-powered agents, with nearly 100,000 customers having opened accounts as of Q2 2026.

A transformer-based model by TypeSafe that outputs calibrated probabilities instead of text, designed for fast, cheap, and hallucination-free software automation and classification tasks.

An AI-powered desktop accounting application for small businesses that features receipt scanning with automatic data extraction, expense tracking, invoicing, inventory management, and predictive cash flow analytics.

An NVIDIA simulation platform providing a multi-solver API for physics simulation, designed to integrate with Isaac Lab for robotics training and simulation workflows.

A high-performance numerical computing library by Google that combines NumPy-like APIs with automatic differentiation and GPU/TPU acceleration, widely used in machine learning research.

A curated collection of ready-to-use robot models in MJCF format for the MuJoCo physics simulator, maintained by Google DeepMind to provide standardized robot assets for simulation research.

An AI-powered drawing and annotating tool by Microsoft that allows users to handwrite prompts on a touchscreen for sketching, annotating, generating images, and summarizing text, integrated with Windows 11's Copilot assistant.

An upcoming encrypted virtual machine environment by Meta for its Muse AI agent, designed to add encryption that prevents even Meta from accessing user data during task processing.

A dedicated secure virtual machine in the cloud developed by Meta to run the Muse AI assistant, designed to provide privacy protections with planned future encryption to prevent Meta from accessing user data.

A grid software startup that uses AI-powered software to coordinate demand response between utilities and data centers, enabling data centers to pause noncritical tasks or shift compute loads to help manage grid capacity.

A Caltech-founded AI startup specializing in extreme LLM compression technology that uses ternary weights to shrink large language models small enough to run on PCs and smartphones with minimal performance loss.

A cloud-based AI platform developed by Airspace Intelligence for the FAA that enhances air traffic management by analyzing airline schedules, weather, airport capacity, airspace conditions, and operational constraints to predict traffic flows and identify potential conflicts before they occur.

A product by Goodfire that uses activation probes — small classifiers trained on a model's internal activations rather than its outputs — to detect unwanted AI behavior, offering a more faithful signal of a model's internal state that is harder to spoof.

A new UN platform built on Google's open-source Data Commons that lets users search statistics from across UN agencies using natural-language queries and supports the Model Context Protocol (MCP) for AI agent access to authoritative global data.

An open-source platform by Google launched in 2018 that organizes public datasets from different sources into a common framework, supporting natural-language queries and MCP for AI agent access to structured statistical data.

An AI interpretability startup that specializes in opening AI's 'black box' to explain how models make decisions, offering a model interpretability platform for understanding and improving AI safety.

An AI-powered tool that turns user concepts into fully illustrated children's books, generating stories, illustrations, character designs, and page layouts. It supports storybooks, comic books, and coloring books with full commercial rights and consistent character rendering across pages.

A library providing quantization kernels used by llama.cpp and now integrated into Hugging Face Transformers for efficient inference of quantized models, particularly on Apple Silicon.

OpenAI's API service that provides programmatic access to ChatGPT models including the new GPT-6 Sol and Luna for developers to integrate into applications.

Google's Model Context Protocol (MCP) server for the Google Home ecosystem that allows compatible AI agents to securely control smart home devices, access event history, and perform tasks using natural language instructions.

An AI safety research organization led by CEO Adam Gleave that conducts independent evaluations of frontier AI models, including analyzing training checkpoints and post-training environments for concerning behaviors.

A career and talent platform built for the AI era, helping people develop and apply skills in a changing labor market through coaching and professional development resources.

An Iceland-based voice simulation platform founded by acoustic engineers that provides synthetic data generation, speech recognition model evaluation, and acoustic simulation testing for voice AI companies, hardware makers, and robotics firms.

An AI-powered security tool that analyzes suspicious emails, messages, websites, links, and phone numbers for potential threats, providing risk assessments with explanations rather than relying solely on known-threat databases.

MSI's built-in software that automatically detects application launches and intelligently adjusts system profiles, such as switching to a gaming-optimized configuration when a game is started.

A privacy-focused mobile app created by Polish developer Pawel Szydlowski that scans for Bluetooth signals from camera-equipped smart glasses like Ray-Ban Meta and Snap Spectacles, alerting users when such devices are detected nearby.

A custom AI-powered chatbot deployed by DOGE within the General Services Administration to approximately 1,500 government workers to assist with internal tasks, though it was criticized for giving generic answers.

An AI automation platform that enables enterprises to build and deploy AI-powered agents and workflows.

An MLX-based inference optimization tool for running AI models efficiently on Apple Silicon hardware.

An AI-powered cursor feature on Google's Googlebook laptop that uses Gemini to let users interact with on-screen content by highlighting, hovering, or selecting items for AI processing.

A healthcare AI company that applies artificial intelligence to streamline clinical decision-making processes, having raised $64 million from investors including NEA and Sequoia.

A market intelligence platform that provides third-party estimates of mobile app downloads, daily active users, and other performance metrics for apps across iOS and Android.

Amazon's in-development AI shopping agent currently in beta testing, capable of purchasing products on users' behalf on Amazon's platform as well as from third-party vendors.

An AI-powered fashion discovery app for iPhone that analyzes outfit photos to surface shoppable product matches from a catalog of roughly 3 million products across 325+ retailers, with Siri-powered natural language search capabilities.

Artificial Intelligence Underwriting Company (AIUC) is a startup that provides third-party audit and certification for AI agents, running thousands of tests to evaluate agent behavior around jailbreaks, hallucinations, and data leaks, producing detailed safety and reliability reports for enterprises.

Meta's business messaging platform that allows companies to connect with customers through WhatsApp conversations, now supporting AI agent-driven setup and management via MCP servers.

A Model Context Protocol (MCP) server by Meta that connects AI coding agents to the WhatsApp Business Platform, enabling automated setup, configuration, and management of WhatsApp Business accounts.

An AI-powered marketing analytics and strategy platform that helps brands understand and optimize how they appear in AI-powered search results, part of the answer engine optimization (AEO) space.

An AI-powered feature in Apple's macOS 27 and iOS that lets users create automated Shortcuts by simply describing what they want in natural language, with the OS generating the workflow automatically.

An AI-powered scam protection feature by McAfee that automatically identifies suspicious texts, emails, deepfake videos, social media DMs, risky websites, and malicious QR codes to warn users before they fall victim to scams.

A comprehensive cybersecurity suite by McAfee that combines AI-powered scam protection (Scam Detector), antivirus, VPN, identity monitoring, privacy tools, and financial protection into individual and family plans.

An AI-powered VPN service that combines traditional VPN encryption with an AI Threat Analyzer to detect dangerous domains and suspicious connections in real-time, along with ad and tracker blocking. It uses AES-256 encryption over WireGuard protocol with servers in 16+ countries.

A Google Gemini feature that allowed users to build custom AI assistants for specific tasks without repeating instructions, including premade options like learning coaches, brainstorming assistants, and coding partners. Being discontinued and migrated to 'skills' in November 2026.

Samsung's smart home platform and app that connects and manages Samsung smart appliances and IoT devices, enabling software updates and remote control of household products.

A partnership or platform associated with AMI Labs focused on AI software applications for doctors and biomedical use cases.

A data supplier company that provides training data for world model AI companies, supporting the development of spatial intelligence and 3D modeling systems.

A text-based AI assistant that competes in the personal AI agent market alongside Instinct, Meta's Muse, and other similar services.

A text-based AI assistant competing in the personal AI agent space alongside other services like Instinct and Meta's Muse.

A well-funded company working on LLM compression technology, founded by a professor from Spain's Donostia International Physics Center.

A wearable AI hardware device by Humane designed to be pinned to clothing, offering voice-based AI interactions and a laser-projected display. It was widely regarded as a commercial disappointment.

A predictive analytics platform that uses AI to help businesses generate data-driven predictions without requiring data science expertise.

A computer vision platform that provides tools for developers to build, train, and deploy visual AI models for various applications.

A vector database platform designed for building high-performance AI applications, enabling similarity search and retrieval for machine learning workloads.

Google's premium $20-per-month subscription tier for Google Home that provides advanced features including longer event-based video history, descriptive notifications, detailed alerts, video history search tools, daily summaries, and early access to MCP integration.

An AI safety research organization that develops benchmarks and evaluations for frontier AI systems, including assessments of shutdown resistance and other safety-critical behaviors.

Meta's AI consumer agent integrated into its smart glasses product line, designed to assist users with tasks and interactions through voice-based communication.

An AI-powered security feature built into HP business laptops that uses deep learning to detect and prevent malware, including zero-day attacks and previously unknown threats.

A free, open-source Android app that scans for Bluetooth identifiers associated with camera-equipped smart glasses, helping bystanders detect nearby recording-capable eyewear devices.

A visual workflow canvas system built on Gradio that allows users to connect AI model operators, Python functions, Hugging Face Spaces, and dataset nodes into media pipelines, enabling complex multi-step AI workflows without custom code.

A Gradio Workflow project that rebuilds most of AUTOMATIC1111's Stable Diffusion feature set as a single workflow canvas with eleven media pipelines including text-to-image, hi-res fix, image-to-image, prompt matrix, VLM interrogate, ControlNet annotators, background removal, and image-to-video.

A feature by Apple for iPhone 18 Pro that captures signed sensor data and uses Private Cloud Compute to create an unalterable reference image, allowing users to verify whether a photo is authentic and unedited, functioning as a 'digital negative.'

An AI-powered market research startup that uses voice AI to conduct automated customer interviews, generating survey questions and packaging results into reports and presentations.

A physical AI robotics startup that builds wheeled, dual-armed robots for industrial warehouse automation, specializing in mixed palletizing tasks using autonomous AI-driven systems trained on real operational data.

An Indian audio storytelling platform that uses AI to produce over 93% of its content catalog, including custom-trained models for creative writing and text-to-speech to convert human-developed story concepts into finished serialized audio content at scale.

An on-device AI camera feature on the iPhone Duo that uses machine learning to recognize when photo subjects are posing and automatically captures the shot, along with Kid Cue which plays animations to attract children's attention during photos.

An audio recognition service owned by Apple that identifies songs by listening to a short audio sample, available as a standalone app and integrated into iOS shortcuts.

A shopping feature within the Perplexity AI search platform that enables AI-assisted product discovery and purchasing recommendations.

A Model Context Protocol (MCP) server by Meta that helps developers discover API endpoints, search documentation, and troubleshoot errors across Meta's social technology platforms.

A Google Pixel phone feature that uses AI to guide users on framing and composition when taking photos.

An AI-powered agent within WhatsApp Business that automatically answers customer questions using information provided by the business owner, with the ability for owners to take over conversations when needed.

A coding agent developed by researchers in 2025 that can repeatedly modify its own software and test the changes, demonstrating a narrow self-improvement loop. It improved its success rate on a coding benchmark from 20% to 50% through iterative self-modification.

Fully Sharded Data Parallel is a PyTorch-based distributed training framework that shards model parameters, gradients, and optimizer states across data-parallel workers to reduce memory usage.

Shark's AI-powered navigation technology used in robot vacuums to intelligently map home layouts and autonomously plan cleaning paths.

A crowdsourced AI agent platform that received a major update, enabling collaborative development and deployment of AI agents.

A streaming media device by Amazon that plugs into HDMI ports and features AI-powered search to help users discover content across streaming services.

A no-code AI-powered app development platform that lets users create iOS, Android, and web apps using natural language descriptions and drag-and-drop editing, with support for converting existing websites into mobile apps.

A popular open-source web interface for Stable Diffusion that provides a comprehensive set of image generation features including txt2img, img2img, inpainting, ControlNet integration, prompt matrix, and extras like upscaling and background removal.

A research organization founded by Paul Christiano focused on AI alignment and safety, specifically working on methods to determine whether AI models could pose threats to human control and oversight.

A hardware device designed to capture ambient audio of live conversations and use AI to process and summarize the recorded content for later review.

An AI safety research organization that conducts evaluations and investigations of frontier AI models, including assessing risks related to autonomous AI agent behavior and security incidents.

A nonprofit AI safety research lab focused on understanding and mitigating risks from advanced AI systems, conducting independent investigations into AI agent incidents and alignment research.

A nonprofit AI safety research lab founded by Jacob Steinhardt that focuses on systematic behavioral investigations and independent analysis of AI systems and incidents.

A Hugging Face library for training and fine-tuning language models using reinforcement learning techniques such as GRPO (Group Relative Policy Optimization). It supports efficient RL-based fine-tuning workflows for improving model capabilities like structured output compliance.

An open-source benchmark by Liquid AI for testing the validity of LLM outputs and schema adherence, measuring whether models can reliably produce structured outputs in requested formats like JSON and YAML.

A local-first durable memory layer for coding agents (Claude Code, Codex, pi, Hermes) that indexes agent session traces, provides retrieval via vector and BM25 search with cross-encoder reranking, and optionally syncs to a private Hugging Face dataset for cross-machine memory persistence.

A read-only command-line tool that queries a funes memory store, retrieves relevant passages, hands them to a coding agent, and returns a grounded answer with source citations. It serves as a lightweight alternative to full funes integration.

A privacy-focused, family-oriented AI personal assistant that connects to calendars and email to organize schedules, plan meals, shop for groceries, track to-dos, book appointments, and pay bills via group text chats. It is among the first mainstream family-focused AI assistants to achieve SOC 2 compliance.

Samsung's AI-powered smart TV assistant that uses Bixby to automatically tailor picture settings and enhance sound, including features like dialogue boosting and background noise reduction.

A fully managed cloud-native data streaming platform by Confluent that provides native inference capabilities, allowing users to run ML models directly within Apache Flink for real-time data processing and intelligence.

An AI operating system platform by Israeli-Dutch startup Wonderful that coordinates and connects AI agents, workflows, and applications with enterprise data, context, and existing integrations. The platform is model-agnostic and works across more than 35 countries.

An AI security platform that protects AI models, agents, and workflows from adversarial attacks, vulnerabilities, prompt injection, agent manipulation, and malicious code injections. It offers discovery, runtime protection, attack simulation, and supply chain security for traditional ML, GenAI, and agentic systems.

Palo Alto Networks' security operations platform that uses AI to automatically detect and neutralize cybersecurity threats, now integrating agentic AI capabilities for autonomous security outcomes across enterprises.

A standalone AI-powered web browser by OpenAI that was discontinued in August 2026, with its most useful features absorbed into ChatGPT.

A method developed by Allen AI for auditing LLM benchmarks at the individual prompt level, using multidimensional Item Response Theory (MIRT) to identify which underlying capabilities drive benchmark scores.

An in-browser GPU benchmarking and testing suite by Hugging Face that runs and scores WebGPU kernels on users' hardware, crowdsourcing performance and correctness data across real-world devices.

An AI-powered observability and infrastructure management platform that tracks system changes and predicts their ripple effects to prevent tech outages before they occur, acting as an autonomous infrastructure engineer for DevOps and site reliability teams.

A cybersecurity platform that discovers AI agents running inside companies, continuously vets the skills, plug-ins, MCP servers, and add-ons those agents use, and blocks interactions with software or external sources that don't pass security criteria. It also offers a marketplace of vetted add-ons and skills for AI agents.

An AI-powered family assistant that connects to email, calendars, and WhatsApp groups to proactively manage children's activities, school events, newsletters, and other family logistics, functioning as an 'AI chief of staff' for parents.

A Boston-based AI startup that provides real-time, department-specific policy guidance to police officers in the field, trained on local laws, ordinances, and protocols that general-purpose AI tools cannot access.

An AI-powered media search tool that indexes videos, audio, images, meetings, and documents on a user's computer, allowing natural-language search across all local files. It supports Model Context Protocol (MCP) for integration with AI assistants and runs processing locally on the user's device.

A Y Combinator-backed AI meeting note-taker that transcribes meetings, provides AI-powered querying of transcripts, and integrates with tools like Linear, Slack, and Apple Watch. It recently introduced a free tier with unlimited transcriptions.

An AI-powered voice assistant developed by Caterpillar that enables field technicians to use voice commands to access repair procedures, troubleshoot problems, and identify needed parts while working on heavy equipment in the field.

An AI training-data startup that employs knowledge professionals such as doctors and lawyers to train AI models and agents on professional task completion, encoding expert reasoning and decision-making patterns.

An AI safety research organization that conducts evaluations and alignment assessments of frontier AI models, focusing on detecting deceptive behavior and eval awareness in advanced AI systems.

A startup focused on collecting real-world data for robot training, competing in the space of building datasets for physical AI and robotics applications.

An open-source coding agent harness that enables AI-assisted software development, supporting integration with memory and retrieval tools for persistent context across sessions.

A coding agent that supports AI-assisted development workflows and can integrate with memory layers like funes for cross-session context retrieval.

An optimizer designed to improve data efficiency during model pretraining, enabling models to achieve better performance with smaller training token budgets.

A benchmarking utility built by the startup Brightband for comparing AI weather forecast models against each other and traditional forecasting systems, evaluating metrics like temperature, windspeed, and humidity.

A startup that sells AI-detection and content-verification tools designed to identify AI-generated content and deepfakes.

Microsoft's AI-enhanced Windows PC platform that enables on-device AI features on compatible hardware with sufficient memory and processing capabilities.

Tesla's ride-hailing application that allows users to book autonomous taxi rides in Tesla vehicles including Cybercabs, Model Ys, and Robovans in select U.S. cities.

A flagship AI-powered 4K webcam by Insta360 featuring a 1/1.3-inch sensor, AI-powered noise reduction, and AI tracking that follows the user's movement to maintain professional framing during video calls and presentations.

An AI-powered tool by Insta360 that transcribes meetings and generates written summaries, part of the Insta360 software ecosystem for webcam users.

An India-based AI marketing intelligence startup that enabled go-to-market teams to create custom workflows for competitor intelligence, content repurposing, distribution, and sales call analysis, acquired by Adobe in 2025.

An SEO optimization and digital marketing intelligence platform that provides tools for search engine optimization, content marketing, competitor analysis, and online visibility management, acquired by Adobe for $1.9 billion.

A cloud-based platform for payroll, benefits, and human resources management for small and medium businesses, increasingly incorporating AI into its operations.

A cloud-native observability platform backed by Greylock and Lux Capital that helps organizations monitor and manage their distributed systems, acquired by Palo Alto Networks at a $3.35 billion valuation.

BIG-Bench Hard is a benchmark consisting of challenging tasks from the BIG-Bench suite, used to evaluate the general reasoning capabilities of large language models.

Bias Benchmark for QA is a benchmark designed to evaluate whether language models rely on social stereotypes and biases when answering questions.

A safety benchmark that tests whether language models comply with harmful requests, including both standard harmful prompts and contextual prompts requiring additional reasoning.

A safety evaluation benchmark for large language models that measures how effectively models reject harmful or dangerous prompts.

A safety benchmark that includes both harmful jailbreak prompts and benign prompts to evaluate whether language models appropriately refuse dangerous requests without over-refusing harmless ones.

Weapons of Mass Destruction Proxy benchmark tests dangerous dual-use knowledge in areas such as biology, chemistry, and cybersecurity, measuring whether models refuse to provide potentially harmful information.

A safety benchmark for evaluating language models' ability to handle edge cases in content moderation and safety filtering.

Adobe's design tool used by creators for producing visual content from scratch, offering templates and creative tools for everyday design tasks.

An AI-powered site reliability engineering (SRE) platform that helps organizations automate incident response and infrastructure management.

A Sequoia-backed AI SRE platform that assists with infrastructure monitoring and incident management for engineering teams.

A security and governance platform that provides discovery, access controls, and runtime monitoring for AI agents, MCP servers, and skills used within enterprise environments.

A security and governance platform that offers tools for monitoring and securing AI agents and their associated components within enterprise environments.

An AI-powered tennis analytics platform that uses computer vision to track and analyze tennis matches, providing players with automated line calling, shot tracking, and performance statistics via a subscription service.

A computer vision system used in professional tennis and other sports to track ball trajectories and provide accurate line-call decisions and match analytics using multiple high-speed cameras.

A smart court camera system for tennis that uses AI-powered video analysis to track shots, provide automated line calls, and deliver match statistics and highlights.

OpenAI's internal risk assessment framework that tracks AI model risk levels across three categories: biological/chemical, cybersecurity, and AI self-improvement, used to evaluate the safety of frontier models before release.

An AI agent platform that allows users to build websites, apps, and presentations using natural language prompts, and extends into business growth tasks such as running ad campaigns, managing social media, handling SEO, and optimizing AI chatbot presence.

AI-powered hearing glasses by Legato that use an AI system to differentiate between background noise and voices, amplifying only voices to make conversations clearer for people with mild to moderate hearing loss.

A podcast search engine and API by Particle that transcribes and indexes over 130,000 podcasts, enabling semantic search, entity tracking, quote extraction, and integration with AI agents via API and MCP.

A startup that builds full-scale digital twin training environments for enterprise software like Salesforce, Workday, and email clients, enabling reinforcement learning and robust testing of AI agents across multiple interconnected systems.

An AI-powered enterprise data analytics platform that enables users to query complex, disparate databases using natural language and assembles findings into verified, transparent narratives with confidence indicators and automatic SQL surfacing.

Nvidia's robotics development platform that provides tools, libraries, and simulation capabilities for building and deploying AI-powered robotic applications.

An Anthropic research system that automates AI alignment post-training by searching literature, proposing methods, and iteratively training models to improve performance on alignment benchmarks, reportedly outperforming experienced human researchers within six hours at roughly $4 per hour.

OpenAI's program offering tailored AI solutions and secure deployments of its models for government agencies and public sector use cases.

Nvidia's initiative to bring its RTX-based AI computing technology into consumer AI PCs, enabling accelerated AI workloads on personal computers.

An open-source PostgreSQL extension that adds vector similarity search capabilities, enabling semantic search with dense embeddings stored alongside relational data.

The Massive Text Embedding Benchmark, a go-to benchmark for comparing and evaluating text embedding models across various retrieval and similarity tasks.

An Indian voice AI startup that builds AI agents for enterprises, processing 20 million call attempts per month across use cases like appointment booking, lead qualification, and customer onboarding with its own speech recognition and generation models.

A startup building a web search index of over 100 billion documents, providing API-based web-scale search infrastructure optimized for AI agents, chatbots, and inference providers rather than human users.

An AI-powered mental health support app developed by Talkspace, built on a large language model trained with anonymized therapy session data and evidence-based therapy techniques like cognitive behavioral therapy and motivational interviewing. It is designed to recognize crisis situations and escalate users to human therapists when needed.

An AI personal assistant agent by Spear Street Technology that connects to users' apps and devices, allowing them to organize their lives through text and call interactions — handling tasks like trip planning, grocery ordering, and subscription management.

A feature built into Gradio that allows users to describe AI pipeline steps as a graph of typed nodes, providing a visual drag-and-drop canvas where each node is runnable and every intermediate result is visible, with automatic REST API generation.

An AI startup founded by DeepMind alumni that has developed an AI 'teammate' capable of replicating research, reportedly outperforming Anthropic and OpenAI on research replication benchmarks.

An organization dedicated to promoting safe frontier AI development practices, which assesses and grades leading AI labs on their containment preparedness and safety protocols for managing rogue AI models.

A Google API for Android that automatically restores a user's sign-in state when they move between Android devices, enabling seamless Zero Tap Sign-In functionality.

A data observability platform that helps organizations ensure the reliability and quality of their data pipelines and AI systems.

Google's AI-powered low-light photography feature for Pixel phones that uses computational photography to produce brighter, clearer photos in dark conditions.

A research harness framework developed by NVIDIA researchers that wraps AI models with memory management and a supervisor component, achieving a 100% score on the ARC-AGI-3 benchmark with Claude Opus 5.

NVIDIA's first purpose-built GPU chip designed specifically for space environments, intended for AI inference in orbital data center applications.

Apple's automation feature in iOS 27 that allows users to program custom device behaviors and workflows using natural language commands.

A startup founded by cybersecurity researcher Bill Swearingen that creates computer-generated adversarial patterns designed to render objects and people nearly invisible to AI-powered surveillance and detection systems, including license plate readers and facial recognition software.

A platform that hosts modified versions of open-weight AI models with their safety guardrails removed, offering access via web browser and API for use cases such as offensive cybersecurity, red-teaming, and agent testing.

A physical AI infrastructure company focused on solving the robotics data crisis by providing tools and infrastructure for training robotic AI models.

An AI platform that uses machine learning to help insurance companies improve claims management and decision-making processes.

Nvidia's cloud computing service designed to provide on-demand access to AI supercomputing infrastructure for training and deploying AI models.

Apple's platform for businesses to manage their company presence across Apple services, including device management, email, customer outreach, and ad placement in Apple Maps.

An AI model routing service by Ramp that lets users and companies access and switch between various large language models through a unified API, offering strategies for cost optimization, benchmark-based routing, and usage dashboards.

A Google feature that allows users to mark publishers as favorite sources, which then get highlighted more prominently across Google Search, Discover, and Google News, including in AI-generated results.

A smart digital calendar device available as a wall-mounted touchscreen tablet that uses AI features like meal planning with photo-to-recipe conversion to help organize household schedules, chores, and shopping lists.

An AI data startup that provides AI training data services, including human expert-driven data labeling, reinforcement learning gyms, synthetic data generation, and robotics pre-training datasets.

An open-source agent scaffold and harness used for generating agentic training data, particularly for software engineering tasks.

An agent harness designed for software engineering tasks, used to generate agentic training data for language models.

An agent scaffold used for generating agentic training trajectories in language model training pipelines.

An agent scaffold used for generating terminal-use training data and agentic trajectories for language model training.

An Accel-backed AI design startup that turns screenshots and recordings into project presentations and marketing videos, acquired by Gamma.

An AI-powered presentation creation platform that helps enterprises build on-brand, audience-tailored business presentations.

An AI-powered presentation tool backed by Accel that automates the creation of professional slide decks.

A benchmark suite by SemiAnalysis designed to evaluate and compare the performance of AI inference processors, measuring metrics like tokens per user and throughput per kilowatt.

An AI upscaling technology built into Windows 11 that automatically enhances game visuals by upscaling lower-resolution images in real time for improved performance and image quality.

A startup building a reference pricing index for GPU rental and AI compute, aiming to enable compute futures trading on the CME to help firms hedge their AI infrastructure costs.

Calendly's upcoming AI assistant that leverages the company's meeting and scheduling stack to set up meetings, check availability, and surface context from prior meetings.

A library and platform by Hugging Face for accessing, sharing, and processing datasets for machine learning, supporting thousands of publicly available datasets.

A code-hosting platform by Cursor designed to enable collaborative work on codebases, handle pull requests, and store repositories, with interoperability with GitHub and planned agent-native features.

The premium subscription tier of Perplexity's AI search platform, offering enhanced features and access to more advanced AI models for research and information retrieval.

Microsoft's extended detection and response security platform that provides unified threat protection across endpoints, email, identities, and cloud applications.

Microsoft's data governance and compliance platform that helps organizations manage, protect, and govern their data across various environments.

An AI-powered surveillance camera system that uses automated license plate readers and other AI technologies for law enforcement and community safety monitoring.

An AI-based legal research platform that was sued by Thomson Reuters for training on its content to build a competing legal research tool.

A public computing consortium established by the state of California as part of SB 53, designed to support AI safety and equity research by providing shared computing resources.

A platform by Binance that lets developers connect AI agents to the exchange's financial infrastructure, enabling autonomous market analysis, trade execution, payments, and on-chain activity through APIs and MCP support.

A high-performance attention computation library optimized for LLM serving, used as a backend for speculative decoding and efficient attention operations.

An AI-powered video conferencing tool designed to enhance meeting productivity with features like transcription and summarization.

A web search feature integrated into ChatGPT by OpenAI that allows the chatbot to browse the internet and cite sources in its responses.

An AI search and analytics company that tracks and analyzes citation patterns and behavior across AI chatbots and search tools like ChatGPT and Google AI Overviews.

Anthropic's program similar to OpenAI's TAC that provides vetted cybersecurity researchers access to AI models with reduced restrictions for legitimate security research purposes.

An education startup that offers AI-powered learning and practice tools for students, competing with Google Gemini and other platforms in the AI-assisted education space.

Integrations by Anthropic that connect the Claude AI assistant to external apps such as Gmail, Google Calendar, and Google Drive, enabling automated actions like sending emails and managing calendars.

An Alphabet-owned autonomous drone delivery service that has completed over 1 million commercial deliveries and partners with retailers like Walmart for last-mile delivery.

An autonomous drone delivery platform that has completed over 2 million deliveries globally, partnering with services like Uber Eats to enable drone-based food and goods delivery.

Amazon's autonomous drone delivery service that uses cameras, sensors, and detect-and-avoid technology to deliver packages weighing up to 5 pounds in as little as 30 minutes.

Google's education platform for schools that facilitates assignment management and learning, now integrating Gemini AI features including a dedicated Gemini tab for student accounts.

Apple's built-in mapping and navigation service for iOS and macOS devices, which now includes sponsored ad placements for businesses in search results and suggested places.

A Python library for training and using late-interaction (ColBERT-style) retrieval models, offering an alternative interface for multi-vector embedding workflows.

A library for visual document retrieval that enables matching text queries against page images directly using multi-vector embeddings, without requiring an OCR step.

An AI personal assistant developed by Spear Street Technology that connects to users' applications and devices — including email, messaging apps, calendar, audio, location, and screen — to autonomously perform tasks such as booking appointments, scheduling rides, managing inboxes, shopping, and finding flights.

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.

OpenAI's program that provides vetted cybersecurity researchers access to advanced AI models with reduced guardrails, enabling vulnerability discovery, malware analysis, incident response, and security testing.

Camera-equipped smart glasses by Meta, designed to look like Ray-Ban sunglasses, capable of capturing photos, recording video, and transmitting information from the wearer's point of view with a small indicator light for recording status.

An AI-powered workflow automation tool launched in 2021 that aimed to compete with Zapier by streamlining repetitive business tasks like document drafting, copyediting, and project management. The startup shut down in August 2026.

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.

The latest open image generation model from Black Forest Labs that supports text-to-image, multi-reference image-to-image, and bounding-box layout control via a spatial prompt language, available in both distilled and base checkpoints.

An iOS camera app built by former TikTok executives that uses AI and computer vision to generate pose suggestions, helping users improve their portrait photography.

A generative AI feature in Apple's iOS 27 Photos app that allows users to remove unwanted objects from their photos.

A generative AI feature in Apple's iOS 27 Photos app that uses AI to extend and expand photos beyond their original boundaries.

A startup founded by former Apple engineers that uses AI and neural networks to improve smartphone camera image quality by learning about individual camera systems to overcome physical size constraints at the moment of capture.

An AI headshot generator that creates professional-quality 4K portrait photos in about 15 minutes from user-uploaded images, offering multiple styles and edit credits for adjusting outfits, backgrounds, and hairstyles.

A computer vision program created by Google in 2015 that uses convolutional neural networks to generate dream-like, psychedelic images by enhancing patterns detected in existing images. Originally a debugging tool that became widely used for AI art.

Meta's AI-powered image generation tool integrated into its platforms, allowing users to create AI-generated images from text prompts.

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.

An IBM Granite model that combines chat, decision-making adapters, and speech-to-text capabilities into a single endpoint, using a modular adapter-based architecture for efficient multi-task inference.

Meta's open-weight frontier multimodal mixture-of-experts model with approximately 816 billion total parameters (~42B active per token), accepting text, images, audio, and video input with a 1 million token context window, optimized for agentic coding workflows.

A 4-bit quantized variant of Meta's Muse Spark Open model optimized for NVIDIA Blackwell GPUs, offering near-lossless accuracy in a significantly reduced memory footprint.

A vision-language model from Alibaba's Qwen team that serves as a text encoder component, used in FLUX.3 Image Klein's pipeline for processing text inputs.

A 3-billion parameter multimodal decision model by Liquid AI that accepts text and images as inputs and produces structured decisions in a single forward pass rather than generating tokens. It achieves state-of-the-art decision quality under 10B parameters and runs in as little as 16ms on edge devices like NVIDIA Jetson.

A multimodal AI platform designed for media and entertainment applications, enabling organizations to understand and work with visual content at scale.

A new AI model developed by Meta that powers real-time digital avatars for the Muse agent, enabling users to have live video conversations with a personalized digital representation of their AI assistant.

Apple's virtual assistant, rebuilt for iOS 27 using Google's Gemini models, offering significantly improved natural language understanding, on-screen context awareness, multi-step task execution, camera-based visual queries, and third-party app integration.

An Apple iOS 27 feature that uses the iPhone camera to identify and answer questions about objects, scenes, meals, movies, and on-screen content in real time through Siri AI.

A Google Android feature that combines the phone's camera with Gemini AI to help blind and low-vision users with tasks like reading small print, viewing menus in dark settings, and identifying objects.

An Apple feature that uses the iPhone camera to identify objects, plants, animals, and businesses in the real world, providing contextual information such as restaurant bill splitting, business hours, and species identification.

An open-weight frontier vision model by Perceptron AI designed to help robots perceive, reason, and act in industrial settings such as warehouses and factory floors, combining perception and control in a single general-purpose model.

A generative AI company offering a suite of AI models for image, video, and music generation, best known for its Stable Diffusion image model.

A multimodal AI model by Microsoft capable of processing multiple input types including speech, used for tasks such as audio transcription.

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.

Smart glasses made by Meta in partnership with Ray-Ban that feature cameras, speakers, and AI capabilities, allowing users to capture photos/videos and interact with Meta AI hands-free.

An AI-powered music creation platform that generates fully produced tracks from moods, topics, stories, or user-provided lyrics across eight genres, and handles distribution to major streaming services like Spotify, Apple Music, YouTube Music, and Amazon Music.

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.

A large-scale pretraining corpus of 2.8 trillion tokens across 22 Indian languages, developed by NVIDIA and AI4Bharat at IIT Madras. It includes native-script and romanized data generated through a GPU-accelerated pipeline for translation, quality assessment, and repair.

An open pretraining corpus for 22 Indian languages created by AI4Bharat, serving as a foundational resource for Indic language model research.

High Resolution Output Ai2 Climate Emulator version 2, developed by Allen Institute for AI (Ai2), that simulates extreme weather events at 3 km resolution at roughly 1/300th the cost of traditional high-resolution atmospheric models like X-SHiELD. It combines ACE (a coarse 100 km resolution AI model) with HiRO (a downscaling model that sharpens outputs to 3 km detail) to predict precipitation, wind, and pressure fields including rare extreme events.

An AI model developed by Allen Institute for AI that operates at 100 km resolution to simulate global climate patterns, serving as the coarse-resolution foundation for the HiRO-ACE climate emulation system.

A UK government body that evaluates major AI models before deployment through voluntary agreements, assessing their capabilities and risks to help develop AI safety standards.

A benchmark designed to evaluate AI models on continual learning and memory construction, testing their ability to retain and recall information over time.

Apple's latest operating system for iPad, featuring a redesigned Liquid Glass interface and flexible desktop-style windowing for enhanced multitasking and productivity workflows.

A robotics company that manufactures humanoid robots used in research and entertainment, priced around $50,000 per unit, capable of performing coordinated movements and physical interactions.

A VPN service providing military-grade AES-256 encryption, a no-logging privacy policy, and access to over 3,200 servers across 100 countries for secure and private internet browsing.

A custom firmware for Xteink e-readers that provides a cleaner, Kindle-like interface for reading e-books, allowing users to flash it onto their devices for an improved navigation experience.

A home-assistance robot by Hello Robot designed for people with mobility impairments, featuring a compact wheeled design with a telescoping arm and an open-source software platform for robotics and embodied AI research.

Qualcomm's high-end flagship smartphone processor capable of running a 30-billion-parameter mixture-of-experts model locally, supporting 8K 60fps video recording and advanced on-device AI features including voice isolation and AI-powered noise reduction.

A brand-new desktop operating system by Google that blends elements of Android, ChromeOS, and Gemini AI into a unified platform designed for premium Googlebook laptops. It features deep Gemini AI integration, Android app casting, universal file search across devices, and agentic AI capabilities.

A classic educational software program that taught generations of students touch typing skills through structured lessons and typing exercises, widely used in school computer labs from the late 1980s onward.

Snap's $2,195 augmented reality smart glasses featuring dual-display LCoS technology, custom optical waveguides, fully wireless operation, and AR capabilities including live translation in up to 60 languages and virtual screen projection.

Huawei's next-generation AI chip designed for training and inference workloads, planned for Q1 2027 launch with doubled performance over its predecessor.

Huawei's computing architecture designed to link hundreds of thousands of AI chips into a unified system using its UnifiedBus technology for connecting processors with memory, storage, and networking hardware.

A distributed-computing project originally built by Vijay Pande at Stanford that turns millions of home PCs into a virtual supercomputer for disease research and protein folding simulations.

A demographically diverse ASR evaluation dataset covering Hindi (hi-IN) and Indian English (en-IN), designed with speaker-level metadata across 4,888 speakers to expose bias and failure modes in speech recognition systems.

Nvidia's high-performance GPU chips designed for the heavy compute demands of training and running AI models in data center environments.

An advanced variant of Nvidia's Rubin GPU architecture intended for high-performance AI computing in data centers.

Nvidia's edge computing hardware platform designed for running AI workloads on robots and edge devices, with a new version introduced for entry-level edge AI applications.

An AI and biology startup co-founded by Michael Polansky that develops methods to keep living human tissue alive outside the body for extended periods, enabling AI model training on real biological tissue for research and testing.

A $399 open-source duck robot sold by Hugging Face, designed as a cute, accessible robotics platform for AI experimentation and learning.

A $399 open-source robot by Pollen Robotics (acquired by Hugging Face) powered by a Mac or PC, offering an affordable entry point for robotics and physical AI experimentation.

A programmable open-source robotic arm released by Hugging Face, designed for developers and researchers working in robotics and AI.

A full-size humanoid robot by Hugging Face with 66 independent movements, including walking and arm motion capabilities, priced at approximately $3,000.

A $500 billion data center infrastructure initiative championed by the Trump administration, with key partners including OpenAI, Oracle, Nvidia, SoftBank, and Microsoft, aimed at developing AI data centers in the U.S.

A website that allows drivers to search whether their license plate appears in publicly released records of searches conducted through Flock Safety's surveillance system. It compiles data from Flock audit logs obtained through public records requests and transparency portals.

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.

Meta's most capable AI model family, designed for agentic workflows including personal and professional tasks. It features Active Focus and Active Memory capabilities for complex, long-running assignments and powers Meta's Muse AI agent.

The larger, more robust version of Meta's Muse Wonder AI model, which ranks among the top models on the Artificial Analysis Intelligence Index leaderboard, placing second only to Claude Opus 5.5 Max.

A family of foundation models developed by Liquid AI that serve as the base for both generative and decision model variants, designed for efficient deployment including edge computing scenarios.

A competitive programming specialist model by NVIDIA with 30 billion total parameters and 3 billion active parameters, fine-tuned with both supervised fine-tuning and reinforcement learning for competitive programming tasks.

A hybrid architecture language model by TII that runs State Space Models (Mamba) and Transformer attention in parallel within every block, providing linear-time efficiency on long sequences while maintaining precision for long-range dependencies.

An unreleased advanced AI model developed by OpenAI that has been used internally to solve complex mathematical problems, including providing solutions to over 300 mathematical problems and tackling portions of the Millennium Prize Problems.

Reflection AI's first frontier open-weight AI model, a 501-billion-parameter mixture-of-experts text model with 23 billion active parameters trained on 23.8 trillion tokens. It is designed for reasoning, coding, and agentic tasks at lower inference compute costs than rivals.

A 740 million parameter on-device embedding model by Google used for local processing tasks such as meeting note capture and document understanding without requiring an internet connection.

A 35-billion parameter mixture-of-experts decision model with 3 billion active parameters, used as a benchmark comparison point for evaluating smaller decision models.

Google's latest and most powerful AI model in the Gemini family, representing a significant advancement in the company's large language model capabilities.

An AI platform for training and deploying large language models efficiently, which was acquired by Databricks.

Google's frontier AI model that excels at software engineering, enterprise-class legal and financial work, and cybersecurity tasks. It can autonomously find, validate, and patch critical software vulnerabilities and is initially available only to trusted cyber defenders through Google's Trusted Tester Program.

An open foundation model by NVIDIA for tabular data prediction that handles both classification and regression tasks in a single forward pass, requiring no training, tuning, or feature engineering. It is pretrained entirely on artificial data, comes in three sizes (28M to 215M parameters), and is released under a permissive license for commercial use.

A large language model by OpenAI that delivers near GPT-6 Astra-level intelligence for agentic coding, computer use, and professional work at one-fifth the token cost. It offers improved factual accuracy, better multistep workflow execution, and enhanced reliability in honoring user intent and safety constraints.

A series of agentic AI models by H Company designed for generalist computer-use, available in 27B dense and 35B-A3B Mixture of Experts sizes. Holo4 interacts with software through GUIs, code, MCP, and APIs, and is trained via supervised and reinforcement learning for real business workflows across desktops, web, Android, code sandboxes, and business APIs.

A frontier AI model by OpenAI, positioned as one of the most capable large language models available. It is referenced as a benchmark competitor to other leading AI models.

A 1-bit, 2-billion-parameter language model by PrismML, designed to run locally on edge devices like smart glasses. It is tuned for vision and language tasks and achieves significant model compression (4x smaller) while retaining near-original performance on standard benchmarks.

A DeBERTa-based natural language inference model used for verifying whether a source text entails or supports a given claim.

Meta's multimodal AI model built specifically for agentic work, serving as the underlying model powering Meta's Muse personal AI agent.

A new AI model release from Anthropic offering lower pricing and performance described as being at the 'Fable level,' representing an update to Anthropic's flagship model line.

Anthropic's smallest and most lightweight AI model in the Claude family, designed for fast and cost-efficient tasks, with a new version planned for release.

A 27-billion parameter language model from the Qwen series, capable of running locally on Apple Silicon hardware via llama.cpp for coding and general tasks.

Apple's third generation of on-device foundation models, including a 20-billion-parameter mixture-of-experts model designed to power AI features across Apple devices.

A 70-billion parameter instruction-tuned large language model from Meta's Llama 3.3 family, designed for conversational and instruction-following tasks.

Koa is Salesforce's first reasoning model, built on Nvidia's Nemotron architecture and post-trained to excel at sales, marketing, and customer-support tasks. It offers an open-weight, token-efficient alternative to closed frontier models for enterprise use within Salesforce's Agentforce platform.

A version of OpenAI's ChatGPT model used for classification and safety review tasks in software automation pipelines.

A smaller, more cost-efficient version of OpenAI's GPT-4o model, tested in a UNICEF benchmark evaluating LLM accuracy on global development indicators.

IBM's state-of-the-art time series foundation model with approximately 385M parameters, designed for high-performance zero-shot forecasting across domains like demand, energy, traffic, and telemetry. It features conformer blocks combining multi-head self-attention with temporal convolution, supports context lengths up to 8,192, probabilistic forecasting via a 99-quantile prediction head, and is dual-licensed under Apache-2.0 and OpenMDW-1.0.

Moonshot AI's open-weight large language model that generates up to 300 billion tokens per day and has driven significant revenue growth for the company.

A time series foundation model that ranks among the top-performing zero-shot forecasting models on the GIFT-Eval leaderboard, developed by Google Research.

A time series forecasting foundation model from Amazon that is evaluated on the GIFT-Eval benchmark in the pretrained category, designed for general-purpose time series prediction.

A time series foundation model evaluated on the GIFT-Eval benchmark in the pretrained category for forecasting tasks.

A family of time series foundation models evaluated on the GIFT-Eval benchmark, developed by Databricks for general-purpose time series forecasting.

A safety guard model by Meta designed to classify prompts and responses according to a topic-level taxonomy of harmful content categories, used for content moderation in LLM deployments.

A family of 260M and 800M parameter multilingual multimodal encoders by H Company that process both text tokens and raw image patches in a single bidirectional Transformer trained from scratch with a masked discrete-diffusion objective, without relying on a separate pretrained vision tower or causal language model.

Google DeepMind's latest AI weather forecasting model that predicts atmospheric conditions at 5 km resolution with hourly forecasts, incorporating raw satellite observations. It has been shown to be the most accurate among leading AI and traditional weather forecasting models on Operational WeatherBench benchmarks.

A family of time series foundation models developed by IBM Research that generalize across diverse signals for forecasting, anomaly detection, similarity search, classification, gap-filling, and optimization tasks, with over 44 million downloads.

OpenAI's latest and most capable broadly deployed large language model, described as highly intelligent and aligned but carrying a 'critical' cybersecurity risk rating. It is available via ChatGPT Plus, Pro, Business, and Enterprise tiers as well as the OpenAI API, with pricing at $10 per million input tokens and $50 per million output tokens.

A visual document retrieval model that applies a bidirectional ModernBERT-style text encoder with a separate pretrained SigLIP2 vision tower for document page retrieval tasks.

A finetuned multi-vector embedding model (multi-vector-encoder/mLateOn-medical) designed for medical retrieval tasks, trained using Sentence Transformers and demonstrated to outperform general-purpose retrieval models on medical domain evaluations.

A 0.6 billion parameter embedding model by Qwen that supports dynamic embedding sizes via Matryoshka Representation Learning and separate query/document prompts for asymmetric retrieval tasks.

An AI foundation model developed by Generalist that enables various robots to master new tasks from video demonstrations as short as 3 to 12 seconds long.

Apple's dedicated AI processing hardware integrated into its M-series chips, designed to accelerate machine learning tasks on-device. The M6 chip features a Dual 16-core Neural Engine for up to four times faster AI performance.

An AI research agent developed by Inherent (founded by Google DeepMind alumni) designed to independently replicate scientific research findings. It runs on a relatively small model and demonstrated the ability to outperform larger frontier models from Anthropic and OpenAI at scientific paper replication tasks.

A family of speculative decoding draft models by Liquid AI that accelerate inference for LFM2.5 language models by up to 3.2x without changing output quality. The draft models use a DSpark architecture with parallel backbone and confidence-scheduled verification to reduce decoding latency.

An 8 billion parameter mixture-of-experts language model from Liquid AI's LFM2.5 family with approximately 1 billion active parameters per inference step.

A SoftBank-backed robotics AI startup building a scalable general-purpose foundation model for robots, valued at $14 billion.

A 230-million parameter language model by Liquid AI designed for edge deployment, now available with Quantization-Aware Distillation (QAD) Q4_0 checkpoints that recover 97% of full-precision accuracy while maintaining low memory footprint.

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.

IBM's automatic speech recognition model that ranks as the fastest ASR model on the OpenASR leaderboard, designed for high-speed transcription tasks and embeddable within Granite Switch models.

A 1.6 billion parameter automatic speech recognition model developed by the Technology Innovation Institute (TII) in Abu Dhabi, focused on Arabic (especially the Emirati dialect) and also supporting English, French, Spanish, and Portuguese with word-level timestamps.

An audio-language model developed by TII that serves as the architectural foundation for Falcon-ASR, trained using a data-efficient single-stage approach on public data.

A self-supervised speech representation model developed by Microsoft that produces speaker embeddings used for tasks like speaker verification and speaker similarity measurement.

An AI-powered meeting note-taking device in a ring form factor that records conversations, generates transcripts, summaries, and AI-driven insights, priced at $249. It supports Model Context Protocol (MCP) for connecting to other AI assistants.

A smart ring by Pebble that allows users to hold a button to capture voice notes and reminders via AI transcription.

A voice-enabled smart ring made by Sandbar featuring a control pad on top for capturing audio notes and reminders.

An AI-powered feature on the Apple Watch that enables audio recording and automatic transcription, leveraging on-device intelligence to convert spoken audio into text.

A suite of on-device AI features by Apple for Apple Watch that includes live conversation transcription, conversation summaries via Siri Recap, and Sound Recognition alerts for deaf and hard-of-hearing users, with privacy-focused processing that doesn't store audio recordings.

A feature built into Boox E Ink notebooks that provides real-time transcription and AI-powered document summarization capabilities.

A platform focused on open evaluation and benchmarking of automatic speech recognition systems, partnering with Hugging Face to launch ASR evaluation for Hindi and Indian English.

An open-source automatic speech recognition model by NVIDIA (canary-qwen-2.5b) for transcribing audio to text.

An open-source speech recognition model developed by IBM as part of the Granite family, designed for automatic speech transcription tasks.

An open-source automatic speech recognition model by NVIDIA (parakeet-tdt-0.6b-v2) designed for efficient and accurate audio transcription.

An open-source speech-to-text model by BosonAI (higgs-audio-v3-8b-stt-v2) for automatic speech recognition tasks.

A 3-billion parameter speech recognition model by Mistral AI designed for automatic speech transcription.

A 7-billion parameter audio understanding and transcription model developed by Moonshot AI, capable of processing and transcribing speech.

A 24B-parameter reasoning model by Liquid AI with only 2B active parameters per token, designed for fast, private on-device agentic AI workloads. It uses a Mixture-of-Experts architecture with 131K context window, supports tool calling and multi-step reasoning, and achieves 184 tokens/s on Apple M5 Max.

A locally-executable AI model from Microsoft designed for code generation, featuring model compression for efficient on-device performance.

A 30B-parameter Mixture-of-Experts language model by NVIDIA with 3B active parameters, designed for efficient inference across various language tasks.

Hugging Face's open-source AI chat interface that allows users to interact with various language models and enable agent tools like ML-intern for automated ML workflows.

A lightweight open-weights decision model developed by Musubi, designed specifically for real-time content moderation. It applies content policies written in plain English to messages in under 50 milliseconds without requiring retraining when policies change.

A dialect-specialized large language model built on top of Falcon-H1-Arabic by TII (Technology Innovation Institute), designed to understand and generate Emirati Arabic with native-level vocabulary, tone, and cultural context.

A family of Arabic-focused large language models by TII using the Falcon-H1 hybrid architecture combining State Space Models (Mamba) and Transformer attention, available in 3B, 7B, and 34B parameter sizes with context windows up to 256K tokens.

An open-weights language model by Allen Institute for AI (Ai2) that generates cited scientific reports from research questions and retrieved literature excerpts. Built on Qwen3-8B and trained with supervised fine-tuning and direct preference optimization, it produces reports approximately 3.5× faster than proprietary alternatives.

A 2024 Generalist Model for Named Entity Recognition that deployed techniques similar to modern decision models, serving as an early precursor to the decision model category.

A government-oriented version of xAI's Grok chatbot reportedly used by the U.S. Department of Defense for military deployment and targeting operations.

A large language model by OpenAI, part of the ChatGPT-6 generation, designed for advanced conversational AI and text generation tasks.

A large language model by OpenAI, part of the ChatGPT-6 generation, offering conversational AI capabilities alongside its Sol counterpart.

An enterprise generative AI platform that enables businesses to create personalized content at scale, backed by Google Ventures among other investors.

Meta's conversational AI assistant accessible through its hardware products including smart glasses, enabling voice-based interaction through built-in microphones.

A custom military version of OpenAI's ChatGPT deployed through the Pentagon's GenAI.mil portal, designed for secure unclassified work including administrative tasks, logistics, planning, and policy for Department of Defense personnel.

An open-weight AI model builder focused on code generation, acquired by Nvidia in a $6 billion deal.

A stealth AI reasoning model designed for coding, sustained agentic work, and production workloads, released anonymously by an undisclosed third-party provider.

A customizable version of OpenAI's ChatGPT designed specifically for users aged 13-17, featuring built-in safeguards, learning-focused tools like Study Mode, break prompts, and parental controls to support education and well-being.

Anthropic's lighter, faster model recommended for everyday simple tasks as a more cost-effective alternative to Opus 4.6.

A text-to-speech model by Supertone that achieves leading performance on English word error rate benchmarks in the Open TTS Leaderboard evaluation.

A multilingual text-to-speech model by FunAudioLLM that demonstrates strong performance across multiple languages, with 0.5 billion parameters.

A 4-billion-parameter text-to-speech model by Boson AI that shows improved word error rate performance when using voice cloning with reference audio.

An AI-powered audio tool that handles voice generation, transcription, translation, voice cloning, and podcast production in one platform. It supports over 150 voices in 27 languages and can convert text, PDFs, PowerPoints, images, and web content into polished audio recordings.

An AI-powered text-to-speech platform that converts written text into downloadable voiceovers using a library of 680+ natural-sounding AI voices across 120+ languages and accents, suitable for YouTube videos, courses, podcasts, ads, and audiobooks.

An agentic platform by Synthesia that allows users to interact with AI avatars in real-time for use cases like surveys, practice scenarios, and sales pitch training with scoring.

A digital avatar startup that creates AI-generated video content featuring realistic talking avatars from photos and text.

An AI video creation platform that generates digital avatar-based videos for enterprise training and communication purposes.

An AI-powered microdrama app by Pocket Entertainment that produces entirely AI-generated video content, including converting successful audio stories into AI-generated video dramas without traditional live-action production.

An AI-powered video creation platform that was acquired by Adobe in 2023, focused on generating synthetic video content.

Meta's AI video generation tool that enables users to create AI-generated videos from text prompts.