Frontier AI
On November 16, Copilot removes Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra from Chat, agent mode, inline edits and completions.
Why it matters. Move workflows to Claude Opus 5.5, GPT-6.1 Sol or GPT-6 Luna. Business and Enterprise admins who turned off default enablement must allow them.
Source: GitHub changelog
Opus 5.5 fast mode needs no access request or beta header, gets default rate limits in Start, Build and Scale tiers, and reaches Microsoft Foundry.
Why it matters. Teams can turn on faster Opus 5.5 output without a waitlist. Fast mode on Opus 5 and Opus 4.8 still needs the beta header.
Source: Claude release notes
In beta for Claude Enterprise, the Compliance API returns each member's Instructions for Claude and Cowork Global instructions through a new user_preferences endpoint.
Why it matters. Compliance teams can audit the standing instructions people give Claude, using a Compliance Access Key with the read:compliance_user_data scope.
Source: Claude Compliance API docs
Web Search on Amazon Bedrock AgentCore now runs in US East (Ohio), US West (Oregon), Asia Pacific (Sydney) and AWS GovCloud (US-West).
Why it matters. Government workloads can now ground agents on the web inside GovCloud, and other teams can keep search closer to their apps.
Source: AWS What's New
Meta's open weight coding model, with 816B total and 42B active parameters and a 1M token context, deploys from JumpStart in BF16 or NVFP4.
Why it matters. Teams can run a large open coding model inside their own AWS account. The NVFP4 variant needs roughly a quarter of the memory.
Source: AWS What's New
GPT-Live sessions can delegate to Decisions directly, or trigger decisions as input or output transcript deltas arrive, using gpt-live-1 and gpt-6-luna.
Why it matters. Voice apps can act while the user is still speaking instead of waiting for the turn to end.
Source: OpenAI API changelog
grok-voice-3.0 is available for Speech to Speech at the Grok Voice 2.0 price, and grok-voice-latest now points to it.
Why it matters. Sessions on the latest alias upgrade automatically. Apps pinned to grok-voice-2.0 should switch, since that model is deprecated.
Source: xAI API release notes
Nemotron 3.5 Super VL is a 120B mixture of experts model, 12B active per token, reading text, images, documents and video.
Why it matters. It ships on Hugging Face, NVIDIA NIM and OpenRouter, giving teams another open option for agents that read documents and screens.
Source: NVIDIA technical blog
Timestral comes in 5M, 28M, 123M and 209M parameter sizes under a modified MIT licence, with a 32,768 step context.
Why it matters. Forecasting teams get open models that run on premises or air gapped, with fine tuning through Mistral Forge.
Source: Mistral news