Claude Opus 5.5 cuts API prices and targets longer coding jobs

01A new Opus, with an efficiency pitch

Anthropic released Claude Opus 5.5 on 22 September. The company highlights codebase-wide changes and extended tasks, and reports fewer tokens per completed job. Those are vendor findings, not a guarantee that every existing automation will improve.

The practical audience is builders whose assistants repeatedly inspect files, call tools and revise their work. In that setting, the cost of one answer tells only part of the story: a task may involve many model calls before the result is usable.

02What the API actually costs

Standard pricing is $4 per million input tokens and $20 per million output tokens. Cache reads cost $0.20 per million. Input and output rates are 20% below Opus 5; cache reads are 60% lower.

Anthropic’s larger claim of roughly 40% lower running costs combines pricing with token efficiency on typical workloads. It is not a flat discount on every request. Cache writes, uncached input and generated output are different billable categories.

For illustration, one million uncached input tokens plus 100,000 output tokens costs $6 at those standard rates, before other charges. The same quantities at Opus 5’s $5/$25 rates cost $7.50. This arithmetic isolates the price change; it does not predict how many tokens a real task will consume.

03Fast mode is a separate tradeoff

Fast mode is offered in Claude Code and the Claude Platform at $8 per million input tokens and $40 per million output tokens. Anthropic advertises up to 2.5 times the speed, so the faster option has a different price calculation.

For an overnight maintenance job, latency may matter less than the total bill. For an interactive coding session, waiting time can matter more. Those are different reasons to choose a mode, even when the underlying model is the same.

04Where it is available

Opus 5.5 is available on Claude Pro, Max, Team and Enterprise, and through the Claude Platform and supported cloud providers. The native API model identifier is claude-opus-5-5.

Our take: the release is worth attention for its economics as much as its benchmark position. A lower rate is easy to compare; better performance on your particular codebase still has to show up in completed, correct work.