Claude Opus 5.5 undercuts Opus 5 at $4 per million tokens, and Pro and Max limits go up
Anthropic released Claude Opus 5.5 on September 22 and priced it below the Opus 5 it steps ahead of: $4 per million input tokens and $20 per million output, against Opus 5's $5 and $25. The company is also raising five-hour usage limits on Pro, Max and Team plans, and giving subscription users a rate limit reset they can save and spend whenever they choose.
Anthropic's launch card for Claude Opus 5.5, published on September 22. Image: Anthropic
Cache reads fall further, from $0.50 to $0.20 per million tokens, a 60% cut to what Anthropic calls the majority of agentic and coding work costs. Five-minute cache writes drop from $6.25 to $5, one-hour writes cost $8, and batch requests stay at half price.
The company's own headline figure is larger than the price list alone supports. "Our tests show that at default settings it will cost 40% less than Opus 5 on typical workloads," Anthropic says, a claim that leans on the model finishing the same job with fewer tokens as well as charging less for each one. Per token, input and output are 20% cheaper.
Fast mode doubles the price for up to 2.5x the speed
A research preview called fast mode runs Opus 5.5 at up to 2.5 times the output speed for $8 per million input tokens and $40 per million output. It is limited to the Claude API, not Amazon Bedrock, Google Cloud or Microsoft Foundry, and access runs through an account manager or a waitlist. At standard speed, Anthropic says the model generates output tokens more than 30% faster than Opus 5.
Claude Code 2.1.280 makes Opus 5.5 its default Opus model, listing the same $4 and $20 pricing and $0.20 cache reads. The API model ID is claude-opus-5-5, and Anthropic says the model is available on Amazon Web Services, Google Cloud and Microsoft Azure as well.
Thinking is always on and the default effort drops to medium
Adaptive thinking cannot be switched off on Opus 5.5, and the effort setting that governs how long it thinks now defaults to medium, where Opus 5 defaults to high. Anthropic says that at medium the new model matched or beat Opus 5 at high on its coding and knowledge-work evaluations, in fewer steps and with fewer tokens.
Carrying over an Opus 5 effort level is the expensive move. At the same setting Opus 5.5 thinks more per turn, especially at xhigh and max, so turns run longer and output tokens climb, and Anthropic's advice is to start at medium and test the levels against your own evaluations. The context window stays at 1M tokens with a 128K maximum output, and the knowledge cutoff is June 2026.
Four things break in code written for Opus 5, according to Anthropic's developer documentation: thinking cannot be disabled, forced tool use returns a 400, thinking blocks no longer carry across models and conversations, and the older computer_20251124 tool is rejected on the Claude API and Google Cloud.
It narrates its own work, and some apps will go quiet
Between tool calls Opus 5.5 now writes short updates on what it found and what it is doing next. Those updates come back as thinking blocks rather than text blocks, and their text is empty at the default display setting, so an application built to stream that running commentary to its users shows nothing through a long agentic turn until it asks for the summaries.
Anthropic also claims a change in tone. The model puts the most important information up front, it says, with less jargon and closer adherence to the writing rules it is given, and it quotes an early tester saying "it writes the way I do".
Nothing was retired to make room
Opus 5 is still active, and so is every Opus back to 4.5, with no new retirement date on any of them. Anthropic commits to keeping Opus 5.5 itself available until at least September 22, 2027, and says Claude Sonnet 5.5 and Claude Haiku 5.5 "will follow in the coming weeks".
For anyone paying for Claude rather than calling the API, the firm change is the higher five-hour limit on Pro, Max and Team and the reset subscribers can bank. Anthropic has not said which subscription plans can select Opus 5.5 in the Claude apps, where it is merging Claude Cowork into chat and adding Claude Docs and Slides.
On performance, Anthropic's own comparison table puts Opus 5.5 ahead of OpenAI's GPT-6 Astra on most of its nine rows, and Artificial Analysis now scores it 58 on its Intelligence Index, above Claude Fable 5.1 and GPT-6 Astra on 53.
Anthropic calls Opus 5.5 the first model in a new Claude 5.5 family, and its first release since the company called for pacing the frontier.
