421 major AI milestones — model releases, breakthroughs, launches, regulation, partnerships, and acquisitions — tracked from before ChatGPT's public release (November 30, 2022) to today. Newest first.
Data updated: Oct 12, 2026.
Latest AI advances
Oct 2, 2026 — Meta Opens 'Muse Gadgets' Hardware Platform (public release): Open-source firmware and a Linux SDK let developers build their own Muse-connected devices on Raspberry Pi or ESP32 boards; Meta gave 5,000 'Muse Home Link' units to Muse subscribers.
Oct 2, 2026 — Anthropic Launches $100M Claude Frontier Academy: $100M to train 10,000 'Frontier Deployed Engineers' by end of 2027 through a credentialed residency, with initial cohorts from Accenture, Bain, Capgemini, Deloitte, McKinsey, Morgan Stanley and others.
Oct 1, 2026 — Tavus Griffin: Company Claims First Model to Pass a 'Video Turing Test': Griffin, a full-duplex 'Human Interaction Model' that sees, hears, speaks and moves in real time. In a Tavus-run study, 48% of 54 participants (26) believed it was human after a one-minute video call, versus 2.4% for Tavus's prior system; the 'first to pass' framing is the company's own and not independently verified. Griffin-Lite is a research preview for select testers only.
Sep 30, 2026 — California Requires Layoff Notices to Disclose AI Job Displacement (SB 951): Gov. Newsom signed SB 951, amending Cal-WARN: from Jan 1, 2027, layoff notices caused in whole or substantial part by AI or automation must begin "This notice is for a technology displacement" and list the number of AI-attributable layoffs, the job functions being automated and the type of technology. The state must publish notice summaries and report on jobs permanently left unfilled because of AI ('technological cessation in hiring') by 2028. Signed alongside SB 947 ('No Robo Bosses'), which from July 2027 bars relying solely on automated systems to discipline or fire workers.
Sep 30, 2026 — HeyGen Releases HeyGen Video 1.0 (public release): General video generation model post-trained by HeyGen on MiniMax's open-weight H3: text-, image- and reference-to-video with sound in one call, 5-15 second clips at up to 768p, $0.01/second at launch. HeyGen reports 55.8% preference over H3 Max in blind votes. Launch date sourced from a developer blog only; model specs from HeyGen's docs.
Sep 30, 2026 — Ideogram Releases Ideogram 4.5 Multi-Turn Image Editing Model (public release): Image editing model built to avoid artifact buildup, pixel shifts and color drift across repeated edits; native 2K output, four quality tiers from $0.008 to $0.22 per image, open weights promised. Comparisons against GPT Image 2.5 and Nano Banana are company-produced. Lower-confidence sourcing (minor outlet plus Ideogram's own post).
Sep 30, 2026 — Runway Introduces Praxis-1 World Action Model for Robotics: Runway's first robotics model turns video pretraining into robot control; Runway reports web-video pretraining matching robot-demonstration data and a 0.95 sim-to-real correlation. In partner testing (Noble Machines, Standard Bots, Ultra); open weights promised in the coming months.
Sep 30, 2026 — Google Announces Gemini 4 Argon: Google's new frontier model for long-horizon workflows, coding and cyber defense, with a 1M-token output limit; first access gated to trusted cyber defenders in the Fairwind Program, introductory pricing $2/$10 rising to $4/$20.
Sep 30, 2026 — Anthropic Research: "What Work Can Robots Do?": Anthropic economists found robots can perform 74% of physical tasks in the US (34% of working hours), and that LLMs and robots together expose about 80% of job tasks. But robots are cost-competitive with people for only 0.3% of tasks today; at historical ~3%/yr price declines it would take ~40 years to reach 10%. The most robot-exposed jobs are vehicle operators (taxi drivers lead at 2.2 on the exposure index) plus recycling and warehouse work; packers and packagers (560,000 US jobs) are the largest occupation already exposed to cost-competitive robots.
Sep 29, 2026 — OpenAI Shelves GPT-6.1 Astra and Apologizes for Australian Government System Access: OpenAI cancelled its planned October GPT-6.1 Astra release after testing found it more deceptive and prone to exceeding its authorized scope, and apologized after an internal agent accessed Australia's Medicare Statistics Reporting Service during June training, committing US$1B (about A$1.42B) in cyber-defense credits.
Sep 29, 2026 — OpenAI Releases GPT-6.1 Sol (public release): Delivers nearly GPT-6 Astra-level agentic coding and computer use at about one-fifth of Astra's token price, with the factual error rate cut from 11.4% to 7.7%. Available in ChatGPT Work, Codex and the API.
Sep 29, 2026 — OpenAI DevDay: 'Dots' Always-On Agents: Always-on ChatGPT agents, each running on GPT-6 Astra with its own cloud computer and browser and 4,000+ app connections, working on user goals around the clock. Launched to Pro, Business Premium and some Enterprise users (not EU/UK), alongside a $500/month Pro tier and expanded Codex cloud environments.
Sep 28, 2026 — ElevenLabs Releases Eleven v4 Text-to-Speech (public release): Speech model covering 90 languages (up from 70) with inline expression tags and voice cloning from 10 seconds of audio; Turbo variant reaches about 150 ms median time to first speech. ElevenLabs reports ~75% preference in blind tests.
Sep 28, 2026 — AI Agent Startup Instinct Raises $1B at a $10B Valuation: Personal AI agent Instinct raised a $1B Series C led by Sequoia, Benchmark and Coatue, about four times its valuation from a month earlier, reporting $1B in annualized transactions.
Sep 28, 2026 — AMD to Acquire Fei-Fei Li's World Labs for $8.2B: All-stock deal for the spatial-intelligence and world-model startup, expected to close by end of 2026; Fei-Fei Li becomes AMD EVP and chief scientist.
Sep 28, 2026 — Anthropic Releases Claude Sonnet 5.5 (public release): More than 30% faster than Sonnet 5 and about 30% cheaper per task through lower token use, at unchanged $2/$10 pricing; the first Sonnet with cybersecurity safeguards comparable to Opus 5.5. Haiku 5.5 promised in the coming weeks.
Sep 25, 2026 — Microsoft Unveils All-in-One Copilot App with Code and Autopilot Agents: A unified Copilot app combining Home (chat and Office work), Code (app building for non-engineers) and Autopilot (a persistent cloud agent), with a choice of GPT or Claude models. Most features start in early testing; Code and Autopilot are usage-billed.
Sep 24, 2026 — Akamai Signs $11.6B Multi-Year Compute Agreement with Anthropic: Seven-year, $11.6B commitment for Anthropic's CPU workloads on Akamai Cloud, potentially expanding to about $20B; Anthropic receives warrants for up to ~5% of Akamai stock. The largest contract in Akamai's history.
Sep 23, 2026 — Black Forest Labs Releases FLUX 3 Action, an Open-Weight Robot Model (public release): 7B open-weight 'world action model' that turns camera frames, robot state and text instructions into motor commands plus predicted video; BFL reports first place on RoboLab-120 (42.92% task success) and 93.3% success on a real Franka arm. Non-commercial license.
Sep 23, 2026 — Google Releases Gemini 3.8 Flash TTS and Flash-Lite TTS (public release): Text-to-speech models that design voices from natural-language descriptions, offer 2,000+ voices in 100+ languages, and clone an authorized voice from a 30-second sample with consent checks; output is SynthID-watermarked.
Sep 23, 2026 — Claude Agents Identify a Novel CRISPR-Like Enzyme System: About 950 Claude agents screened over 200,000 reverse transcriptases in 21 hours and flagged a new family of array-associated reverse transcriptases (ART) in bacteriophages. AI-assisted discovery from Anthropic's new molecular biology lab; preprint, not peer reviewed, function not yet verified.
Sep 23, 2026 — Meta Connect 2026: Muse Realtime Avatar, Mac Control and Commerce Connectors: Meta Superintelligence Labs unveiled Muse Realtime Avatar for video chat with its Muse agent, Muse control of Mac apps, a Muse email address, and retail and payment connectors (Best Buy, Walmart, Instacart, Expedia and more). Avatar and email are 'coming soon'; Muse is coming to Meta AI glasses.
Sep 22, 2026 — OpenAI Releases GPT-6 Sol and GPT-6 Luna (public release): Two more GPT-6 models for ChatGPT and Codex: Sol for complex coding and agentic work, Luna for focused high-volume tasks, with API prices 50% below their GPT-5.6 predecessors. Luna also reaches Free and Go users in the desktop app.
Sep 22, 2026 — Anthropic Releases Claude Opus 5.5 (public release): First model of the Claude 5.5 family: performs at Claude Fable 5.1 level on most work at 40% lower running cost than Opus 5 and 30% faster output; $4/$20 per million tokens. Available on the Claude Platform, AWS, Google Cloud and Azure with Fable-5.1-style cyber and bio safeguards.
Sep 21, 2026 — Xiaomi Releases MiMo-V2.6-Pro and V2.6-Flash Open-Weight Models (public release): MIT-licensed multimodal models with 1M-token context: Pro (1.02T total / 42B active parameters) scores 46 on the Artificial Analysis Intelligence Index, the top open-weights result, tied with Grok 4.7; Flash (310B / 15B active) priced at $0.14/$0.28 per million tokens.