Claude Opus 5.5: Pricing, Context Window and Breaking Changes

What Claude Opus 5.5 is

Claude Opus 5.5 is Anthropic's flagship model, released on 22 September 2026 as claude-opus-5-5. Anthropic positions it for long-running agentic coding and knowledge work rather than for chat. It takes text and image input, returns text, and carries a 1 million token context window with up to 128,000 tokens of output in a single response. It is available on the Claude API and through Amazon Bedrock, Google Cloud and Microsoft Foundry. Thinking cannot be turned off on this model; see the breaking changes below.

Pricing

Input is $4 per million tokens and output is $20, both 20% under Opus 5's $5 and $25. The number Anthropic leads with, though, is a 40% reduction in the cost of typical workloads at default settings, and that is a different claim from the rate cut. Their reasoning is that Opus 5.5 finishes work in fewer turns and fewer tokens, so the bill falls faster than the per-token price does. Cache reads are $0.20 per million tokens, 60% below Opus 5. Batch processing takes another 50% off, at $2 and $10 per million. Anthropic also reports output generating more than 30% faster than Opus 5.

Benchmarks, and who ran them

On LMArena's text leaderboard Claude Opus 5.5 has 1507 Elo (rank 4, high effort); by category, 1547 Elo in Coding (rank 6, high effort), 1514 Elo in Math (rank 8, high effort), 1545 Elo in Expert (rank 4, high effort) and 1517 Elo in Creative Writing (rank 2, high effort). Anthropic's own published results are their claims, not independent measurement: they report that on their internal agentic coding tasks Opus 5.5 matched Opus 5's quality in roughly half the turns, time and output tokens, cutting that workload's cost by 40 to 50%. Their Terminal-Bench 4.0 figure is reported at xhigh effort, and Anthropic notes the standard error is ±2.6 points. Unless stated otherwise the results use adaptive thinking at max effort, and production safeguards were enabled during evaluation, which Anthropic says likely depressed its own scores on cybersecurity and biology tasks.

Four breaking changes if you are migrating from Opus 5

Code running against Opus 5 hits four incompatibilities on Opus 5.5: thinking cannot be disabled, forced tool use returns an error, thinking blocks are now tied to both the model and the conversation, and on the Claude API and Google Cloud the earlier computer_20251124 computer-use tool is no longer accepted. The first three also apply to Claude Fable 5.1. A fifth change breaks nothing: text between tool calls now arrives inside thinking blocks whose text is empty at the default display setting. If your application streams that text to users as progress updates, it will simply go quiet between tool calls until you set a display value that returns it.

Context window

The 1 million token window is roughly 750,000 words. Anthropic position Opus 5.5 for long-running agentic coding and knowledge work.

Cache writes, fast mode and lifecycle

A 5-minute cache write costs $5 per million tokens and a 1-hour cache write $8; Anthropic's pricing documentation states a cache hit on Opus 5.5 costs 5% of the standard input price, where most Claude models charge 10%. Fast mode, in research preview in Claude Code and on the Claude Platform, runs at up to 2.5 times the speed and costs $8 per million input tokens and $40 output. On the Message Batches API the model supports up to 300,000 output tokens with the output-300k-2026-03-24 beta header, and the minimum cacheable prompt length is 512 tokens. The documentation gives a reliable knowledge cutoff of June 2026 and retirement not sooner than 22 September 2027. Anthropic also state that Opus 5.5 was tested before release by external evaluators including METR, and that vetted organisations can apply to a Life Sciences Verification Program to use it for biology research.

Official Resources