Claude Opus 5.5

// Overview

Claude Opus 5.5 is the first model in Anthropic's Claude 5.5 family, released on September 22, 2026 and built for long-running agentic coding and knowledge work. Anthropic positions it as performing at the level of Claude Fable 5.1 on most work while costing 40% less to run than Claude Opus 5.

The model has a 1 million-token context window and up to 128K output tokens on the synchronous Messages API, rising to 300K on the Message Batches API with the output-300k-2026-03-24 beta header. Modalities are text and image in, text out. Reliable knowledge runs through June 2026. Adaptive thinking is always on and cannot be disabled; depth is controlled with the effort parameter, which defaults to medium.

Pricing is $4 per million input tokens and $20 per million output tokens, against $10/$50 for Claude Fable 5.1 and $2/$10 for Claude Sonnet 5. Cache reads cost $0.20 per million and a 5-minute cache write $5 per million; Batch API requests take 50% off both sides. A separately priced fast mode is available on the Claude API as a research preview.

Four changes break code written against Claude Opus 5: thinking cannot be disabled, forced tool use returns an error, thinking blocks are tied to the model and conversation that produced them, and the earlier computer_20251124 computer use tool is not accepted on the Claude API or Google Cloud. A fifth change alters the response shape without failing requests — text between tool calls returns inside thinking blocks whose text is empty at the default display setting.

// Benchmarks

1. Terminal-Bench 4.0 (agentic coding)66.4%
2. FrontierCode v1.154.4%
3. CursorBench 4.0 (agentic coding)57.8%
4. Humanity's Last Exam67.7%
5. Chartography89%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Cache reads $0.20/MTok, 5m cache write $5.00/MTok, 1h cache write $8.00/MTok. Batch API takes 50% off input and output.

Pricing source ↗

// Strengths

• Frontier-level agentic coding at $4/$20 per MTok — 40% cheaper to run than Claude Opus 5
• Leads Anthropic's published comparisons on Terminal-Bench 4.0 (66.4%), FrontierCode v1.1 (54.4%) and Humanity's Last Exam (67.7%)
• 1M-token context with 128K max output, extendable to 300K on the Batch API
• Anthropic reports output generated more than 30% faster and fewer tokens used per task than Opus 5
• Anthropic reports its highest alignment audit scores, with more resistance to prompt injection and less tendency to take hard-to-reverse actions

// Best for

• Reach for it when an agent has to run for hours across a large codebase and per-token cost, not capability, is the binding constraint.
• Reach for it for computer-use and automation workflows — Anthropic reports 81.8% partial on OSWorld 2.0 against 74.0% for Opus 5.
• Reach for it for complex knowledge work and research where long-horizon planning and self-verification matter more than latency.

// How to access

// Claude Opus — every version

The full lineage of the Claude Opus line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.