How much does Claude Opus 5.5 cost?
As reported in our videos, Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, with cache reads at $0.20 per million tokens. Check Anthropic's pricing page for the current rates.
Claude Opus 5.5
Claude Opus 5.5 pricing: what it costs and what changed
Claude Opus 5.5, released September 22, 2026, costs $4 per million input tokens and $20 per million output tokens, 20% less than Opus 5, with cache reads at $0.20 per million tokens. Anthropic's claim is Fable 5.1-level work at 40% lower cost to run than Opus 5.
Updated September 30, 2026 · By Dr. Sam Sammane
From the TheoSym channel
Claude Opus 5.5: pricing, benchmarks and what changed
Published September 23, 2026
A seven-minute briefing on the Opus 5.5 release: token pricing and cache read savings, coding benchmarks, safety audit nuances, four breaking API changes and the questions to ask before you switch.
Watch on YouTubeClaude Opus 5.5 price per million tokens
| Item | Opus 5.5 |
|---|---|
| Input tokens | $4 per 1M |
| Output tokens | $20 per 1M |
| Cache reads | $0.20 per 1M |
| Versus Opus 5 | 20% lower price per token |
Figures are as reported in our September 2026 videos, citing Anthropic. Confirm current rates on Anthropic's pricing page.
Where the 40% saving comes from
The 40% figure is a cost-to-run claim, not just a lower price tag. It combines a lower price per token with fewer tokens spent per task, so the same job costs less to finish.
Cache reads matter most for agents. Re-reading the same context on every step is most of an agent's bill, and cache reads at $0.20 per million tokens make that repeat reading cheap. Our video walks through the cache read savings in its pricing chapter.
What else changed in Opus 5.5
Price is only part of the release. The full briefing covers:
- Speed improvements and a fast mode
- Coding results on Terminal-Bench 4.0
- A research report accuracy test and a more direct writing style
- Safety audit nuances, including misalignment findings and a fallback catch
- Four breaking changes on the API
- A silent change to the default effort setting (medium vs high)
Why Opus 5.5 arrived so quickly
Opus 5.5 came roughly 60 days after Opus 5. In our second video we look at the developer pushback that followed Opus 5, where developers criticized verbosity, overengineering and reliability, and at what the fast turnaround says about the race between Anthropic and OpenAI.
Before you switch: a short checklist
- Read the four API breaking changes and confirm your calls still work.
- Check the default effort setting so your bill does not change silently.
- Re-run your own evals. Synthetic benchmark scores no longer tell the full story.
- Measure cost per finished task and your cache hit rate, not only the price per token.
- Compare against Sonnet 5.5 for the tasks that do not need Opus.
The fine print matters. Anthropic's own caveats about its benchmarks are worth reading, including that the model can suspect it is being tested. Run our model migration cost estimator and prompt caching savings calculator with your own usage.
Claude Opus pricing: common questions
Is Opus 5.5 cheaper than Opus 5?
Yes. The price per token is 20% lower than Opus 5, and Anthropic says the total cost to run the same work is 40% lower once fewer tokens per task are counted.
What are cache reads and why do they matter?
A cache read is when the model reuses context it has already processed instead of reading it again at the full input price. Agents repeat the same context on every step, so cache reads make up most of an agent's bill.
Should I upgrade from Opus 5 to Opus 5.5?
Check the four breaking API changes and the default effort change first, then re-run your own evals. If your tasks pass and cost per task drops, upgrade.
Is Opus or Sonnet the better value?
It depends on the task. Sonnet 5.5 costs about half as much and scored higher on Terminal-Bench 4.0, while Opus 5.5 leads on FrontierCode and CursorBench. See our Claude Opus vs Sonnet comparison.
Want an agent you can trust in production?
TheoSym ships production agents with the eval suite, MCP tools and harness included. Bring one real workflow to a 15-minute call with Sam and see what building it would involve.