What's new in AI What the AI companies, governments and labs announced, each with the date, what it changes for you, the official page and the day we checked it. We add entries and never rewrite them: if one turns out wrong, it gets a correction.
All fields 34 Models 9 Open-weight models 3 Agents and coding 5 Image, voice and video 2 Laws and rules 6 Science 2 Health 2 Law and finance 2 Education 2 Robots 1
34 entries shown
October 2026 2026-10-13 · Models · Google
Google released Gemini 4 Argon in its API and in the Gemini app for Google AI Pro and Ultra subscribers. Google AI Pro and Ultra subscribers can pick Argon in the Gemini app and AI Mode in Search; developers can use it through the API in AI Studio.
2026-10-12 · Models · Meta
Meta released Muse Wonder through its Meta Model API and Muse Code, and Muse Wonder Full for Muse Code subscribers. Developers can call Muse Wonder through the Meta Model API; Meta says both models are built for personal and professional agentic tasks.
2026-10-06 · Models · Mistral AI
Mistral AI launched a public preview of Mistral Large 4 and said it would release the model's weights by the end of October. The weights were not yet downloadable at launch; Mistral says they will come with more details on the architecture and benchmarks.
September 2026 2026-09-29 · Models · OpenAI
OpenAI released GPT-6.1 Sol and says it nearly matches GPT-6 Astra on coding, computer use and professional work at one-fifth of the price. ChatGPT Plus, Pro, Business, Enterprise and Edu users can choose it in ChatGPT Work and Codex.
2026-09-22 · Models · Anthropic
Anthropic released Claude Opus 5.5 and says it performs at the level of Fable 5.1 on most work and costs 40% less to run than Opus 5. Opus 5.5 is offered on all platforms that carry Claude models, including Amazon Web Services, Google Cloud and Microsoft Azure.
2026-09-21 · Models · SpaceXSI
SpaceXSI released Grok 4.7, its model for coding, agentic tasks and knowledge work, on the xAI API and in Cursor and Grok Build. Developers can call it on the xAI API as grok-4.7, and people using Cursor or Grok Build can choose it there.
2026-09-17 · Law and finance · OpenAI
OpenAI introduced Astra for Law, a foundation for law firms and legal technology companies to build AI products and workflows on. Selected law firms get it first through Trusted Access in ChatGPT and Codex; OpenAI says API access is coming soon.
2026-09-15 · Image, voice and video · Google
Google introduced Gemini 3.8 Live and 3.8 Live Extended Thinking, which it calls its most advanced live dialogue models yet. Google says the models make talking to Gemini by voice in the Gemini app, Google Workspace and Search more fluid.
2026-09-10 · Law and finance · OpenAI
OpenAI launched ChatGPT for Financial Services, which includes data from Daloopa, PitchBook, LSEG News and Crunchbase. Eligible financial institutions can get it from OpenAI, with the listed data sets included.
2026-09-10 · Open-weight models · DeepSeek
DeepSeek released DeepSeek-V4.1-Flash, a 552B-parameter mixture-of-experts model, with its weights under the MIT License. Developers can download the weights from Hugging Face and use them under the MIT License, or call the model on the DeepSeek API.
2026-09-10 · Agents and coding · OpenAI
OpenAI released the Agents API in public beta, letting developers build and run cloud agents on the Codex harness, managed by OpenAI. Developers can build agents through an API that uses the same harness and infrastructure as OpenAI's Codex, without hosting that harness themselves.
2026-09-08 · Science · Google DeepMind
Google DeepMind launched AlphaGenome Atlas, with predicted effects of all 9 billion possible single-letter DNA changes in the human genome. Researchers can use it for non-commercial work on DeepMind's website now; commercial use on Google Cloud is due soon.
2026-09-08 · Science · OpenAI
OpenAI said its system's Lean-formalized proof resolves the Navier–Stokes Millennium Prize problem; Clay said it was apparently settled. This is OpenAI's own claim; the Clay Mathematics Institute says its prize process is deliberately unhurried and that it will provide updates.
2026-09-03 · Models · OpenAI
OpenAI introduced GPT-6 Astra, which it calls the world's most intelligent and aligned model, first for a limited set of organizations. OpenAI said ChatGPT Plus, Pro, Business and Enterprise users would get Astra within days of the launch.
2026-09-01 · Health · OpenAI
OpenAI added an Epic health-record integration to ChatGPT for Healthcare, bringing authorized patient context into ChatGPT. It is for healthcare organisations using ChatGPT for Healthcare; the health-record integration is not available for individual accounts.
August 2026 2026-08-28 · Agents and coding · OpenAI
OpenAI said it will wind down its contract supplying OpenAI models to Cursor, with a proposed shutoff date of November 12, 2026. Cursor users who rely on OpenAI models could lose them after the proposed November 12, 2026 shutoff; OpenAI says Cursor will not get its future models.
2026-08-26 · Open-weight models · Qwen
Qwen released the weights of Qwen3.8-Flash-Next, a 125B-parameter multimodal model that previews the architecture of Qwen4. Developers and researchers can try the design Qwen says Qwen4 will use, in a model that activates 6B parameters per token.
2026-08-18 · Education · OpenAI
OpenAI introduced ChatGPT for Teens, a version of ChatGPT designed to help teens learn, with age-appropriate safeguards for users under 18. ChatGPT users under 18 get safeguards meant to reduce exposure to harmful content, in an experience designed to help them learn.
2026-08-14 · Agents and coding · Cursor
Cursor said its acquisition by SpaceX is complete, ending a process that began in April with a partnership with SpaceXAI. Cursor is now part of SpaceX and says access to the world's largest GPU fleet will let it build stronger models that are cheaper to run.
2026-08-02 · Laws and rules · European Commission
The EU AI Act became applicable on 2 August 2026; rules for high-risk uses such as education and employment start 2 December 2027. The Act now applies, with some exceptions; AI used in areas such as education, employment or border control faces the high-risk rules from December 2027.
July 2026 2026-07-30 · Robots · Google DeepMind
Google DeepMind said Gemini Robotics 2 can control full humanoid robots, from feet to fingertips, as well as other two-armed robots. So far the robot-control models, including the one that drives humanoids, are available to early-access partners.
2026-07-27 · Laws and rules · European Commission
The EU's AI Omnibus took effect, extending AI Act timelines; rules for AI in products such as lifts or toys now apply from 2 August 2028. Makers of products such as lifts or toys that contain AI get until 2 August 2028 before the AI Act's rules for those systems apply.
2026-07-27 · Open-weight models · Moonshot AI
Moonshot AI released the full Kimi K3 model weights under the Kimi K3 License. Researchers and developers can download Kimi K3 from Hugging Face and deploy or adapt it within the terms of the Kimi K3 License.
2026-07-23 · Health · OpenAI
OpenAI began rolling out Health in ChatGPT in the US, letting users connect Apple Health and supported medical records. Logged-in US users aged 18 and over, on web and iOS and on the Free, Go, Plus and Pro plans, can connect their records so ChatGPT can explain them in context.
2026-07-15 · Laws and rules · Cyberspace Administration of China
China's interim rules on human-like AI interaction services took effect; providers may not offer minors virtual partners or relatives. Five national agencies issued the rules, in force from 15 July 2026; providers of human-like AI interaction services may not offer minors virtual intimate relationships.
2026-07-14 · Education · Anthropic
Anthropic introduced Claude for Teachers, giving verified US K-12 educators free access to premium Claude features and teaching skills. Verified US K-12 teachers can use it free; those who sign up by June 30, 2027 get a full year of access.
2026-07-09 · Models · OpenAI
OpenAI released GPT-5.6 in three tiers, Sol, Terra and Luna, across ChatGPT, Codex and the OpenAI API. GPT-5.6 started rolling out to ChatGPT, Codex and the API on launch day, as a flagship tier, Sol, and two lower-cost tiers, Terra and Luna.
2026-07-08 · Image, voice and video · OpenAI
OpenAI began rolling out GPT-Live voice models to ChatGPT users worldwide and says they can listen and speak at the same time. ChatGPT users around the world are getting two versions, GPT-Live-1 and GPT-Live-1 mini, as OpenAI rolls them out.
June 2026 2026-06-30 · Laws and rules · Anthropic
Anthropic said the export controls on Fable 5 and Mythos 5 had been lifted and Fable 5 would return to users globally on July 1. From July 1, Fable 5 was to be available again to users globally in Claude.ai, Claude Code, Claude Cowork and on the Claude Platform.
2026-06-24 · Agents and coding · Google DeepMind
Google DeepMind built computer use into Gemini 3.5 Flash, so developers can build agents that act across browser, mobile and desktop. Developers and businesses can use it through the Gemini API and Gemini Enterprise Agent Platform to build agents that operate software on screen.
2026-06-18 · Agents and coding · Google
Gemini CLI stopped serving free users and Google AI Pro and Ultra subscribers on June 18, 2026; Google replaced it with Antigravity CLI. Free, Google AI Pro and Ultra users of Gemini CLI are pointed to Antigravity CLI; the Gemini Code Assist IDE extensions stopped serving them too.
2026-06-12 · Laws and rules · Anthropic
Anthropic said a US export control directive barred foreign nationals from Fable 5 and Mythos 5, so it disabled both for all customers. Fable 5 and Mythos 5 were switched off for every customer, not only for foreign nationals, while Anthropic complied with the order.
2026-06-09 · Models · Anthropic
Anthropic released Claude Fable 5 and restricted Claude Mythos 5, which has cyber safeguards lifted, to its Glasswing partners. Developers on the Claude API and Enterprise customers on consumption-based plans could use Fable 5 from launch day; Mythos 5 was not open to the public.
2026-06-02 · Laws and rules · The White House
A US executive order made it policy to work with the private sector on AI innovation and security and to harden systems against threats. The order's section on frontier models says it creates no mandatory licence or pre-approval for developing or releasing new AI models.
How we pick and check entries: our rules . Found a mistake? Report a correction .