Claude Opus 5.5

What is disclosed

Anthropic released Claude Opus 5.5 on 22 September 2026, the first model of its Claude 5.5 family.¹¹1Anthropic, Introducing Claude Opus 5.5, 22 September 2026. https://www.anthropic.com/claude-opus-5-5 Its size and architecture are not published. The launch post says only that Opus 5.5 needs less compute to serve than Opus 5.¹¹

The context window is 1M tokens. One response can run to 128K tokens, or to 300K on the Message Batches API with a beta header.²²2Anthropic, Claude Opus 5.5 model page, Claude Platform docs, read 29 September 2026. https://platform.claude.com/docs/en/models/opus-5-5/overview Input can be text or images, and output is text. Anthropic puts the reliable knowledge cutoff at June 2026.²²

Thinking is adaptive and cannot be switched off. The effort parameter decides how much the model reasons, and it defaults to medium.²²

What is new

Input costs $4 per million tokens and output $20, both 20% below Opus 5.³³3Anthropic, Claude API pricing, read 29 September 2026: Opus 5.5 at $4 input, $20 output and $0.20 for cache reads; Opus 5 at $5, $25 and $0.50; fast mode on Opus 5.5 at $8 and $40. https://platform.claude.com/docs/en/about-claude/pricing. Arithmetic: 4 / 5 = 20 / 25 = 0.8, so 20% below Opus 5. Cache reads dropped from $0.50 to $0.20 per million.³³ At default settings Anthropic expects typical workloads to cost 40% less, because the model also spends fewer tokens per task.¹¹ Output arrives more than 30% faster than on Opus 5.¹¹

Anthropic's launch table puts Opus 5.5 at 66.4% on Terminal-Bench 4.0, at xhigh effort. In Anthropic's own run, Opus 5 scored 52.3%.¹¹ GDPval-AA v2.1 tests work across 44 occupations. There Opus 5.5 reaches 1846 Elo at max effort, with Claude Fable 5.1 at 1735 and Opus 5 at 1708.¹¹

In the launch post, Opus 5.5 "performs at the level of Claude Fable 5.1 on most work", and in Anthropic's own use the gap between the two is narrower than the scores suggest.¹¹

Anthropic judges the model comparable to Claude Mythos 5.1 in biology and cybersecurity. It ships with safeguards similar to those on Fable 5.1.¹¹ The post also reports signs that the model "often suspects it is being evaluated".¹¹

Using it

On the Claude API the id is claude-opus-5-5, and Amazon Bedrock adds the prefix anthropic. to it.²² Batch requests cost half. Writing to the cache costs $5 per million tokens for five minutes and $8 for one hour.²²

Fast mode runs the same model at up to 2.5 times the speed.¹¹ It is a research preview, sold on the Claude API only, at $8 per million input tokens and $40 per million output tokens.³³

Some requests that worked on Opus 5 now fail. Disabling thinking returns an error, and so does forced tool use.²² Thinking blocks stay tied to the model and the conversation that produced them.²²

The model will stay available until at least 22 September 2027.²² Claude Sonnet 5.5, one tier down, costs half as much per token.⁴⁴4Anthropic, Claude API pricing, read 29 September 2026: Sonnet 5.5 at $2 input and $10 output. https://platform.claude.com/docs/en/about-claude/pricing. Arithmetic: 2 / 4 = 10 / 20 = 0.5.