Claude Opus 5.5

What Anthropic says. The choosing a model guide states that "most workloads start with Claude Opus 5.5" and recommends a capability-first approach: implement with Opus 5.5, optimize prompts, then lower effort or move to a cheaper model once the workflow is tuned. Its selection matrix lists multihour autonomous coding agents, large-scale refactoring, complex systems engineering, vision-heavy workflows and computer use as example uses. The same guide says to move to Fable 5.1 only if evals at xhigh or max effort still fall short on demanding reasoning or long-horizon agentic work.

Our editorial judgment. Opus 5.5 is a sensible baseline for a new project because Anthropic's own routing advice is built around it: you can tune effort down for cost or step up to Fable 5.1 for headroom. For simple chat, extraction or classification at volume, the documented price gap to Sonnet 5.5 and Haiku 5.5 is large enough that we would not default to Opus without test evidence. If you are choosing a coding assistant rather than a raw API model, see our AI coding tools guide.

• Thinking cannot be turned off. The model page says adaptive thinking is always on; depth is controlled with the effort parameter instead.
• Forced tool use errors. Requests with tool_choice of type any or tool return an error, per the breaking changes Anthropic lists for code already running on Claude Opus 5.
• Thinking blocks are tied to the model and the conversation that produced them.
• Computer use version. On the Claude API and Google Cloud, the earlier computer_20251124 tool is not accepted.
• Quiet streams between tool calls. Text between tool calls now comes back in thinking blocks whose text is empty at the default display setting, so apps that stream progress must set a display value that returns it.
• Fast mode limits. Fast mode runs on the Claude API and on Azure-hosted Microsoft Foundry deployments, not on Claude Platform on AWS, partner clouds or the Batch API.
• US-only inference costs more. Setting inference_geo to "us" applies a 1.1x multiplier to all token prices.
• Tokenizer. Anthropic says Claude 4.7 and later models produce about 30% more tokens for the same text, so compare cost per task.

All four current models share a 1M token context window, 128K max output, text and image input and a June 2026 reliable knowledge cutoff, according to the models overview.

The pricing page also lists Claude Opus 5 at USD 5 input and USD 25 output per 1M tokens, so Opus 5.5 is listed at a lower token price than its predecessor.

1. Collect 20 to 50 real tasks with known good outcomes, including long agentic runs and a few that earlier models failed.
2. Run them on Opus 5.5 at medium and high effort, and on Sonnet 5.5, with identical prompts and tools, scoring pass or fail against your own rubric.
3. Record input, cached and output tokens per task, then multiply by list prices to get cost per completed task.
4. If tasks still fail at xhigh or max, test Fable 5.1 on only those task types, as Anthropic's guide suggests.

We read Anthropic's models overview, Opus 5.5 model page, pricing page, choosing a model guide and deprecations page on 15 October 2026. We have not run Claude Opus 5.5, and nothing here is a test result or ranking. Check the vendor pages before committing budget. See our ranking method for how HumanRank scores tools and the model directory for other profiles.

What is Claude Opus 5.5? It is Anthropic's Opus model built, in the vendor's words, for long-running agentic coding and knowledge work, and the model its docs suggest most workloads start with.

How much does Claude Opus 5.5 cost? It costs USD 4 per 1M input tokens and USD 20 per 1M output tokens at list price on 15 October 2026, with cache hits at USD 0.20 per 1M tokens.

What is the Claude Opus 5.5 context window? It is 1M tokens, with up to 128K output tokens, or 300K on the Message Batches API with the output-300k-2026-03-24 beta header.

Can I turn off thinking on Claude Opus 5.5? No, adaptive thinking is always on; you control its depth with the effort parameter, which defaults to medium.

Does Claude Opus 5.5 support batch processing? Yes, the Batch API halves the price to USD 2 input and USD 10 output per 1M tokens.

Should I use Opus 5.5 or Fable 5.1? Start with Opus 5.5 unless your own evals at high effort fall short, because Anthropic positions Fable 5.1 as the step up and lists it at 2.5 times the token price.