Claude Opus 4.7

Two breaking changes. (1) Manual extended thinking via `thinking: {type:'enabled', budget_tokens:N}` is removed — switch to `reasoning.effort` (low/medium/high/xhigh/max). (2) The tokenizer changed; the same prompt may consume 1.00x to 1.35x more tokens. Re-baseline cost and quota dashboards before flipping production traffic.

Anthropic's guidance: xhigh for advanced coding and complex agent loops (default in Claude Code), high for most intelligence-sensitive workloads, max only when evals show measurable headroom past xhigh. max often overthinks — most 'Opus 4.7 feels slow' reports trace back to running max by reflex.

Up to 2,576 px on the long edge, about 3.75 megapixels — more than 3x prior Claude models. This is a model-level change applied automatically; no API flag required. XBOW reports a 98.5% visual-acuity benchmark (vs 54.5% on Opus 4.6).

Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, GitHub Copilot (Pro+, Business, Enterprise) — and through TheRouter on the OpenAI-compatible /v1/chat/completions surface as model id 'anthropic/claude-opus-4.7'.

Not publicly disclosed in Anthropic's launch materials or model documentation as of 2026-05-22. For time-sensitive facts, attach explicit context or route through a retrieval layer.