GitHub is gradually rolling out xAI’s Grok 4.7 in Copilot Pro through Enterprise, including VS Code, JetBrains, Copilot CLI, and the cloud agent. It bills at provider list pricing under usage-based billing; Business/Enterprise admins can gate it via model policy.
Xiaomi published MiMo V2.6 Pro (1.02T MoE / 42B active) and Flash (309B / 15B active) on Hugging Face with text/image/video/audio inputs and a 1M token context, plus RL training resources. Vendor figures say Pro rose from 58.4 to 72.57 on DeepSWE v1.1 after one mixed coding/agent/visual/cyber RL run; independent replication is still pending…
Linear’s Sept 21 engineering post says agents now write most of its tests ( 2,000/week) and the suite nearly quadrupled in 2026. Faster runners, tsgo, less setup, and Vitest shard tweaks cut runner time per test roughly in half and trimmed PR wait from 6 minutes to just over 5…
Mintlify Pro+ docs sites now auto expose five WebMCP tools ( search docs , read page , get site overview , read skill , navigate page ) so browser agents can query and read docs as structured calls instead of scraping the UI. Chrome plans default WebMCP support in Chrome 157 (early November)…
Claude Code 2.1.278 (Sept 19) defaults auto mode safety classification to a free server side check for Claude API, Enterprise, Bedrock, Vertex, Foundry, and gateway sessions when reachable; billed local fallback still applies if a proxy strips the fields. Build 2.1.277 also reads AGENTS.md when no CLAUDE.md exists (not yet on Bedrock, Vertex, or Foundry)…
Four consumers filed a Northern District of California class action alleging Anthropic, OpenAI, Google, and SpaceXAI coordinated to slow frontier AI development after Amodei’s Sept 12 pacing essay. The proposed class covers US paid ChatGPT, Claude, Grok, and Gemini subscribers seeking treble damages; the allegations are unproven…
Spotted something we missed? Start a thread.
