Frontier AI
As of October 2, Copilot deprecates Claude Opus 4.7, Gemini 3.5 Flash, Gemini 3.6 Flash and Kimi K2.7 Code in chat, edits, agents and completions.
Why it matters. Switch to the suggested replacements, such as Claude Opus 5.5 or Gemini 3.8 Flash, which Enterprise admins may need to enable in model policies.
Source: GitHub changelog
Request a Copilot code review through the REST or GraphQL API and set its effort level per request, on Pro, Pro+, Max, Business and Enterprise.
Why it matters. Teams can start reviews from their own scripts, CI and internal tools instead of the pull request page.
Source: GitHub changelog
Non-interactive sessions and subagents now continue from the partial response when an API call times out, instead of failing the turn.
Why it matters. Upgrade if you run Claude Code unattended in CI or through the SDK, where one timeout used to lose the whole turn.
Source: Claude Code v2.1.288 release
Server.connect() now rejects while the server is already connected, and a stateless Streamable HTTP transport serves only one request.
Why it matters. Servers that share one instance across HTTP requests fail from the second request, so build the server and transport inside the handler.
Clef (27B) and Clef Flash (9B) answer structured questions about text and images through the new /v1/systemone endpoint.
Why it matters. They suit scored checks on screenshots or text, and Ollama lists them as decision only so clients do not offer them for chat.
Source: Ollama v0.35.1 release
Query Ceramic.ai, Exa or Linkup through one REST endpoint or a Workers binding, billed from AI Gateway credits at partner list prices.
Why it matters. Agents can ground answers in live results without a separate account for each search provider, and you can bring your own key.
Source: Cloudflare blog