All stories

NVIDIA says Jensen Huang resigned immediately as president, CEO, and board member; CFO Colette Kress became interim CEO while the board begins a permanent CEO search.

Anthropic released Claude Fable 5.2, its active Fable model for demanding coding and long-horizon agent work, with a 1M-token context, 128K maximum output, and lower list pricing than Fable 5.1.

Google launched the Gemini Decisions API, using Gemini Decision 1 for closed-set multimodal decisions across text, images, audio, video, and PDFs. The API can evaluate up to 1,000 yes/no, choice, or rubric questions in one request and returns probabilities and confidence scores.

Google released Gemini Omni 1.2 Flash and made it generally available in the Gemini API and Gemini for business. The video model can natively generate and edit clips up to 15 seconds, while 720p is priced at $0.07 per second, which Google says is 30% lower than Omni 1.1.

OpenAI released Images 2.6 across ChatGPT, Codex, and the API. The new image model adds GIF generation, SVG export, image markup, and a complexity control.

OpenAI says that beginning in December, ChatGPT and Codex usage on Plus, Pro, and Business plans will share one allowance measured in tokens; Instant chat will not use it.

OpenAI says it will honor a common robots.txt signal through which website operators can express preferences about use of their content for foundational generative AI training. OpenAI says it plans to implement the signal in coming months and that its existing crawler controls will remain in place.

Mistral has released Mistral Large 4 open weights. The 1T-parameter, 52B-active multimodal MoE can now be downloaded and self-hosted instead of used only through an API preview.

Google announced Gemini 4 Carbon, a frontier model for long-horizon engineering and cyber defense. It starts with selected defenders and trusted partners before any broader developer, business, or consumer availability.

Google DeepMind released Genie 4, a world model it says can generate real-time interactive worlds with native audio, physics, and human movement, and steer scenes through natural language. The model begins as a research preview for trusted testers.

OpenAI released GPT-6.1 Terra with the API identifier gpt-6.1-terra. The model targets visual reasoning, computer use, and high-volume professional work; OpenAI also published standard API pricing and availability in ChatGPT Work and Codex.

Dream Labs released Dream-1, a public World Action Model API for compatible bimanual, parallel-gripper robots. The company says it can attempt control on unseen robots, tasks, and environments; these are first-party claims that teams should evaluate on their own hardware.

Kimi released K3.1 Pro and its two-to-three-times-faster Highspeed variant for Kimi, Kimi Work, Kimi Code, and the API; the company says full model weights will arrive by November 4.

OpenAI released Codex CLI 0.164.0 with reusable prompt and slash-command presets, enterprise MCP authentication, and credential masking for remote Bash and Zsh sessions, alongside session-resume and sandbox-check fixes.

Anthropic released Claude Code 2.1.309 with fixes for shell startup files escaping the sandbox, mutable MCP configuration, and several permission and worktree-isolation bypasses.

Anthropic released Claude Code 2.1.307 with an API-version migration command, tool-call safety telemetry, and fixes for duplicate HTTP MCP calls and resumed MCP connections.

Anthropic released Claude Code 2.1.305 with a retry path when Vertex, Bedrock, or a gateway rejects tool search, plus fixes for bare-repository worktrees and MCP configuration replacement.

MiniMax has released M3.1 Flash through its API, M Plan and MiniMax Code, listing flat input, output and cached-input token prices; weights are planned for a later release.

OpenCode says a security breach tied to a large-scale distillation attack affected 1.4% of Console users, specifically a subset with external providers linked.

Reflection released Apache-2.0 Beam 1.0 weights, a 501B-total, 23B-active MoE for coding and long-horizon tool workflows.