Anthropic released Claude Opus 4.6 today (February 5, 2026). We need to add this model to all necessary providers in Roo Code.

Announcement: https://www.anthropic.com/news/claude-opus-4-6

Model Details

Key New Features & Changes

• Adaptive thinking (thinking: {type: "adaptive"}) is the recommended thinking mode for Opus 4.6, replacing thinking: {type: "enabled", budget_tokens: N} which is deprecated
• Effort parameter GA -- now generally available (no beta header required), with a new max effort level. Default is high.
• 128K max output tokens (doubled from 64K on previous Opus models)
• Compaction API (beta) -- server-side context compaction for long-running agentic tasks
• Data residency controls -- new inference_geo parameter for US-only inference (1.1x pricing)
• Fine-grained tool streaming (GA)

Breaking Changes on Opus 4.6

• Prefill removal -- Opus 4.6 does NOT support assistant message prefill
• Tool parameter quoting -- changes to how tool parameters are quoted
• budget_tokens deprecated -- thinking: {type: "enabled", budget_tokens: N} still works but will be removed in a future release
• interleaved-thinking-2025-05-14 beta header deprecated -- safely ignored if included, no longer required (adaptive thinking enables interleaved thinking automatically)
• output_format deprecated

References

• Anthropic announcement
• What's new in Claude 4.6
• Migration guide
• Models overview
• Pricing