Released Oct 8, the Functionize MCP server lets Claude Code, Cursor, GitHub Copilot, and other MCP clients ask Functionize Studio in plain language to build and run a smoke or regression suite against the real app, then get the failing step back in the same chat. It's available now to Studio customers and trial users, with a Claude Code plugin that adds skills for Studio.
Codex lead Tibo's Day 4 post (Oct 8) says mid task steering now takes effect instantly, so the model reacts to your corrections right away instead of wasting effort on the old direction, while OpenAI Developers says Ultrafast for GPT 6.1 Sol is rolling out today in the API, Codex, and ChatGPT Work at up to 8x Sol Standard's speed. Astra's Ultrafast has been limited to Pro $500 and eligible Enterprise and Edu plans and bills above Standard, so check your plan and the API pricing page before switching it on…
Announced Oct 8 at Gemini at Work 2026, it handles knowledge work, media, and writing and running code from one prompt box, with @Gemini in Gmail and Docs, access from the command line, Slack, and Microsoft 365, plus an API to run it headless inside other apps. It can spin up sub agents and route to Gemini or Anthropic's Claude models, but it's in private preview for select Workspace Business and Enterprise plans with no firm rollout date…
Released Oct 8 on Hugging Face with the same 2.5B active architecture as Mellum2, the model was post trained with reinforcement learning across millions of sandboxed runs so it can explore a codebase, edit files, and check its own changes, and JetBrains says it serves almost twice as many tokens as Qwen3.5 9B under heavy load. GGUF builds for llama.cpp, Ollama, and LM Studio, plus the multi token prediction head for vLLM, are listed as coming soon…
Announced Oct 6 with Genesys, Instinct, Rocket, Shopify, Stripe, and Walmart, the open standard lets a personal agent start as a guest on a company's site, then sign in with read only or write access the user picks, and finish the job through the website, MCP or OpenAPI endpoints, or the company's own agent in one OAuth session. The v0.1 spec and a reference implementation are due later in October, so nothing is implementable yet…
Announced Oct 7 and due by end of October in Copilot CLI, the Copilot app, and VS Code, a 53GB quantized build of the 137B parameter (6.8B active) MoE scores 70.8% on SWE Bench Verified vs 72.6% for the cloud version, with Auto mode picking local or cloud per task or you selecting it via Windows ML or any OpenAI compatible local endpoint. Microsoft's reference machine is an RTX Spark Surface Laptop Ultra with 128GB unified memory (75.5GB peak at 256K context), and local inference alone doesn't make a session offline…
Spotted something we missed? Start a thread.