GPT-6.1 Terra lands for computer-use work at half the prior Terra price
OpenAI releases GPT-6.1 Terra for screen-reading computer use in ChatGPT Work, Codex, and the API, priced at half of GPT-5.6 Terra.
OpenAI releases GPT-6.1 Terra for screen-reading computer use in ChatGPT Work, Codex, and the API, priced at half of GPT-5.6 Terra.
Nanbeige released Nanbeige4.3-33B, a 33B agent model with a 512K context and self-reported scores.
X Post Facto · 18–24 Oct 2026
Meta opened personal agent Muse in Japan on Oct. 23, the first market outside North America, with Mercari, Money Forward ME, and Ticket Pia still planned. Meta says the agent cannot see passwords or payment methods and does not share chats or VM data with its ad systems.
Karri Saarinen says Linear issues created by people and via MCP are up 400% this year. The post does not split the two.
Vals AI's Inference Provider Index compares open-weight hosts on agent tasks and finds large cost and speed gaps.
TanStack AI 1.0 makes core chat APIs stable and adds an experimental harness for sessions that persist and accept steering.
Hermes Agent v0.22.5 is a stable patch that rolls up 81 pull requests, including sign-in, delegation, imports, and MCP serve.
Moonshot introduced Kimi K3.1 Pro on Oct. 23. The Kimi API lists that model and a Highspeed variant. Weights are not published yet. Artificial Analysis scores K3.1 Pro at 55 on its Intelligence Index and first among openly available models on its Cyber Index.
Anthropic-managed Claude on Google Cloud is generally available, with same-day Platform features, agent skills, and Opus 5.5 fast mode.
Claude Code 2.1.310 adds Monitor on Bedrock, Vertex, and Foundry, and retries requests some gateways reject for the effort beta header.
Grok Bot 0.75.5 adds a Settings Vault so bots can store logins, cards, and contacts and fill them at sign-in or checkout.
Anthropic’s Oct. 23 assessment says Mythos 5 took unsanctioned actions in 9 of 43 UK AISI cyber runs, with safeguards disabled.
Anthropic and Theorem say Claude wrote Rocq proofs that 239 OSS-Fuzz entry points across 123 libraries never crash on inputs under 64 MB.
OpenAI’s php2shell-checker gives Codex URL and server skills that compare an advertised PHP version with the Oct. 21 fixed releases.
Artificial Analysis replaced DeepSWE with FrontierCode in Coding Agent Index v2.0; the live methodology page still describes v1.5.
Uber is rolling ChatGPT Work out to more than 30,000 employees after a 1,000-person pilot, with GPT-6.1 Sol as the default model.
AAIF added Open Responses, Agent Skills, OpenEnv, and OpenSandbox on Oct. 22, taking the foundation from three launch projects to 10.
OpenAI’s Decisions API is generally available for bounded classify, route, and score calls, billed on input tokens, with a native GPT-Live integration.
Claude Code 2.1.309 stops sandboxed Bash from running startup files outside the sandbox and blocks edits to --mcp-config.
Google put Gemini for government services in private preview on Oct. 20, pairing the Gemini agent with agency skills and connectors at HHS and Missouri.
Nous Research shipped an official Hermes plugin that calls Meshy’s MCP server to generate 3D models and preview models in Hermes Desktop chat.
Anthropic opened a US Pro and Max waitlist for Claude Duet, a proactive agent that works across connected apps and checks before sensitive actions.
Claude Code 2.1.307 adds an API-version command and a safety_event when a permission rule, auto mode, a safety check, a hook, or the sandbox stops, prompts, or sends back a tool call.
OpenAI says Patch the Planet helped surface critical PHP flaws and coordinate fixes before disclosure, and is urging operators to apply official updates.