Claude Opus 5.5: what Anthropic's new model changes for an SMB
Anthropic released Claude Opus 5.5 on September 22, 2026. According to Anthropic, it costs about 40% less than Opus 5 on a typical workload and writes more clearly.
3 min read
The SynAI team
Key points
- Anthropic released Claude Opus 5.5 on September 22, 2026, on Claude's Pro, Max, Team and Enterprise plans and through its API.
- According to Anthropic, Claude Opus 5.5 costs about 40% less than Opus 5 on a typical workload, and it writes more clearly.
- An agent in production should move to Claude Opus 5.5 only after a test, because its API has four breaking changes for code written for Opus 5.
What did Anthropic announce?
Anthropic announced a new model, Claude Opus 5.5, available from September 22, 2026 on Claude's Pro, Max, Team and Enterprise plans, through the Claude API, and on Amazon Bedrock, Google Cloud and Microsoft Foundry.
| API price | Opus 5 | Opus 5.5 |
|---|---|---|
| Text sent to the model | $5 | $4 |
| Text produced by the model | $25 | $20 |
| Context re-read from the cache | $0.50 | $0.20 |
API price
Text sent to the model
- Opus 5
- $5
- Opus 5.5
- $4
Text produced by the model
- Opus 5
- $25
- Opus 5.5
- $20
Context re-read from the cache
- Opus 5
- $0.50
- Opus 5.5
- $0.20
Prices in US dollars per million tokens, as of September 22, 2026.
What does it change for an SMB?
For an SMB, Opus 5.5 mainly changes the cost of agents and the clarity of their writing.
- An agent that works in steps costs less. At each step, an agent re-reads its instructions and the work done so far, usually from the cache. According to Anthropic, these re-reads make up most of the cost of agent tasks, and their price drops by 60%.
- An agent's emails should be clearer. According to Anthropic, the model puts the most important information first, follows your writing rules and is harder than Opus 5 to trick with malicious instructions hidden in what it reads. Keep human approval before anything is sent.
- For a voice agent, a faster model is still the better choice. The documentation rates the response speed of Opus 5.5 as moderate, behind Sonnet 5 and Haiku 4.5. Anthropic says Sonnet 5.5 and Haiku 5.5 will follow in the coming weeks.
Should you switch now?
Not for an agent in production, not before a test. According to the documentation, the Opus 5.5 API has four breaking changes for code written for Opus 5, and its default thinking effort drops from "high" to "medium". Anthropic measured the 40% saving at default settings: your test will give you the real figure for your agent. In the Claude app, you can try it today on a paid plan.
- Run one recurring task on Opus 5.5 for one or two weeks, alongside the current agent.
- Compare the quality of the output and the real cost on your volume.
- If both hold, switch the agent and watch the first bill.
Frequently asked questions
- Is Opus 5.5 included in the free plan?
- No. According to Anthropic's pricing page, Opus models come with the Pro, Max, Team and Enterprise plans, not the free plan.
- What is a token?
- It is the unit the API uses to count and bill text: a word, part of a word or a character. According to Anthropic's documentation, a token is about 3.5 characters of English.
