On September 22, Anthropic introduced Claude Opus 5.5. The company positions it for demanding coding and knowledge work, claiming performance close to Fable 5.1 on most tasks. It estimates costs 40% below Opus 5 on typical workloads at default settings. That is Anthropic’s assessment, not a guarantee for every request.
The announcement raises two different questions for users: whether the result improves and how payment changes. The second deserves its own explanation, because API rates, a monthly subscription and the cost of a completed task measure different things.
What has become cheaper
According to Claude Platform pricing, the base US-dollar rates per million tokens are:
| Charge | Opus 5.5 | Opus 5 |
|---|---|---|
| Input tokens | $4 | $5 |
| Output tokens | $20 | $25 |
| Cache reads | $0.20 | $0.50 |
An arithmetic example: one million regular input tokens and 100,000 output tokens cost $6 instead of $7.50, a 20% reduction. This excludes caching, tools and other charges. Fast mode has separate, higher rates: $8 per million input tokens and $40 per million output tokens.
This is why a headline percentage should not simply be applied to an entire bill. Compare the full cost of completing the same task: answer length, retries and review time matter too. A short, unsuccessful answer can be cheap without solving your problem.
What this means for a Claude subscription
Anthropic treats paid Claude plans and API/Console access separately. A chat subscription does not include API credits. The table therefore does not imply an automatic reduction in the monthly price of Pro, Max, Team or Enterprise. In the app, your plan’s features matter; in a custom integration, its separate Console usage matters.
If you use Claude for writing or documents, first check the model available in your account and your current limits. A change in developer pricing is not, by itself, a reason to move to a more expensive subscription.
A detail for API users
The Opus 5.5 documentation specifies the model ID claude-opus-5-5. Adaptive thinking is always on, with reasoning depth controlled through effort; disabling thinking is unsupported. Fast mode remains a research preview on the direct Claude API only, without availability through partner cloud platforms.
Anthropic’s migration guide details breaking changes for the Messages API, including a 400 error for the old setting that disables thinking. Start a custom integration upgrade by reviewing that guide and trying a small set of familiar requests before changing the main workflow.
How to assess usefulness for your work
- Choose two or three tasks with known requirements: edit a text, explain a document or fix a small piece of code.
- Compare how well all requirements are met, the number of factual errors and the manual editing needed.
- For API use, record the cost per completed task including retries; for chat, measure time to a usable result.
- Avoid granting unnecessary access to work data just to try a new model. A public or fictional example is enough to begin.
This is a suggested comparison method, not our own Opus 5.5 test result. For broader selection criteria, see our ChatGPT, Gemini and Claude comparison. Our article on developing future models explains how Anthropic measures AI involvement in its own work.
Join the conversation
Stay on topic and respect other readers. Your first comment may appear after editorial review.