Anthropic · Claude Opus
Claude Opus 5.5
Anthropic's current model for long-running agentic coding and knowledge work, released 22 September 2026. Anthropic says it performs at Claude Fable 5.1's level on most work while costing 40% less than Opus 5. Thinking is always on, so the effort setting, not a thinking switch, is how you control depth and cost.
Start here
- What this helps you do
- Decide whether Claude Opus 5.5 is the right model for a job you have, or one to leave alone.
- What you leave with
- A plain read of what Claude Opus 5.5 is for, what to avoid giving it, and the facts behind that.
- What to check before trusting it
- Adaptive thinking is always on and cannot be disabled, unlike Opus 5.
This is the current Opus, so a migration from Opus 5 is a real decision rather than a version bump. Test the four breaking changes first, especially thinking no longer being disableable, then re-baseline cost against the 40% headline saving on your own task rather than Anthropic's framing.
Use it for
- Long-running agentic coding and long-horizon tasks
- Complex knowledge work and multi-step reasoning
- Migrating an existing Opus 5 or Opus 4.8 workload to a cheaper current model
Four changes break code written for Opus 5: thinking can no longer 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. It is also poor value where Sonnet 5 already passes the task.
Effort without the jargon
The work is routine enough that a shallower pass holds quality at a fraction of the tokens and latency.
This is the default on the Claude API, and the sensible starting point before tuning in either direction.
The task is genuinely demanding and you can give it a large max_tokens; note that thinking cannot be disabled at these levels.
Advanced recordCost, limits, privacy and operations
Opus 5.5 is cheaper than Opus 5 on the same headline task and Anthropic says it matches Fable 5.1 on most work, but thinking can no longer be disabled, so measure total billable tokens on your own task rather than reading the lower rate card as a lower bill.
- API cost
- Officially disclosedClaude API list price: US$4 per million input tokens and US$20 per million output tokens, both lower than Opus 5's US$5/US$25. Cache reads are US$0.20 per million tokens against Opus 5's US$0.50; a 5-minute cache write is US$5 and a 1-hour write is US$8 per million tokens; the Batch API gives a 50% discount on input and output.
- Context window
- Officially disclosed1,000,000 tokens
- Maximum output
- Officially disclosed128,000 tokens (synchronous Messages API, 300,000 on the Message Batches API with the output-300k-2026-03-24 beta header)
- Knowledge cutoff
- Officially disclosedReliable knowledge: June 2026; training data: June 2026
- Recorded inputs
- Text · Images
- Recorded output
- Text
Data handling is a product decision
Depends on surfaceAnthropic says commercial API inputs and outputs are deleted from its backend within 30 days by default, with exceptions for stored services, agreed controls, safety enforcement and law. Consumer Claude and cloud platforms have separate terms.
Operational constraints
- Adaptive thinking is always on and cannot be disabled, unlike Opus 5.
- Forced tool use is not supported, unlike Opus 5.
- Thinking blocks are tied to the model and the conversation that produced them.
- On the Claude API and Google Cloud, the earlier computer_20251124 computer-use tool is not accepted.
- Text returned between tool calls arrives in thinking blocks that are empty at the default display setting, so a streaming integration that shows that text as progress updates goes quiet until display is set to return it.
- Effort defaults to medium, lower than Opus 5's high default.
Record details
- Evidence readiness
- informational
- Guidance confidence
- moderate
- Evidence revision
- claude-opus-55@v1
Unknown means the cited official record does not disclose a safe value. It is not an estimate. Prices are provider list prices in USD where stated and can change before this record's review date.
Keep the evidence separate
Anthropic says Claude Opus 5.5 performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5, at US$4 input and US$20 output per million tokens with cache reads at US$0.20 per million tokens (against Opus 5's US$5, US$25 and US$0.50). It has a one-million-token context window, 128k max output, defaults to medium effort and thinks adaptively at all times.
No sufficiently useful independent note has been added yet. That absence is not filled with our guess.
No Human Bit result is claimed; use this as evaluation guidance only.
Not scheduled
No Human Bit result is claimed for this model yet.