Comparison · Chat & code
Claude Opus 5.5 vs Claude Sonnet 5
Claude Opus 5.5 vs Sonnet 5 for coding in Claude Code: same 1M window, different rates and thinking rules. When Sonnet is enough and when to pay for Opus.
What each one accepts
| Claude Opus 5.5 | Claude Sonnet 5 | |
|---|---|---|
| Vendor | Anthropic | Anthropic |
| Type | Chat & code | Chat & code |
| Released | 2026-09 | 2026-06 |
| Context window | 1,000,000 tokens | 1,000,000 tokens |
| Max output | 128,000 tokens | 128,000 tokens |
| Endpoints | /v1/chat/completions · /v1/messages | /v1/chat/completions · /v1/messages |
Token prices
| Claude Opus 5.5 | Claude Sonnet 5 | |
|---|---|---|
| Input / 1M tokens | $3.78 | $3.24 |
| Output / 1M tokens | $18.90 | $16.20 |
| Input from cache / 1M | $0.189 | same as input |
| 1,000 replies, 1,500 in / 400 out | $13.23 | $11.34 |
Charged from a prepaid USD balance, no taxes added. One key runs both, through /v1/chat/completions in the OpenAI format. A failed generation is $0.
Sonnet 5 handles most coding work well and costs less per token than Opus 5.5. Both read up to 1M tokens and write up to 128K, so the decision is about how hard the task is, not about limits. I call in Opus 5.5 for wide migrations, whole-repository reading and agents that run for hours alone.
When to pick Claude Opus 5.5
- You are moving a project between frameworks, for example Flask to FastAPI, and the change touches hundreds of files.
- A Claude Code agent should keep working on its own for a long session.
- You want one request to take in a full repository, a long spec or a stack of contracts.
- You need a planner: Opus splits the job into steps, and a cheaper model writes the code.
When to pick Claude Sonnet 5
- It is your everyday model in Claude Code for tests, bug fixes, small features and review.
- You are hunting the root cause of a bug in old, messy code.
- An agent with a handful of tools runs many times a day.
- You produce docs, reports and emails in bulk.
Differences you will notice
- Rate. Claude Opus 5.5 is billed at $3.78 / $18.90 per 1M tokens, Claude Sonnet 5 at $3.24 / $16.20 per 1M tokens. Opus is the more expensive one on both input and output. The table above also shows whether each has a cached input rate.
- Class. Anthropic places Opus 5.5 at Fable 5.1 level on most tasks, and Sonnet 5 near Opus 4.8. That is a step in tier, not a whole generation.
- Thinking. On Opus 5.5 adaptive thinking cannot be switched off. On Sonnet 5 it is on by default. Either way, reasoning tokens are billed as output, so short answers to hard questions can cost more than they look.
- Sampling fields. Sonnet 5 returns a 400 error if you send a non-default
temperature,top_portop_k. Delete those fields when you port older code. - Age. Sonnet 5 shipped in June 2026. Opus 5.5 arrived on September 22, 2026, with knowledge up to June 2026.
How to decide
- Most of your tasks are routine edits - set Sonnet 5 as
ANTHROPIC_MODELand leave it there. - A task is risky or spans the whole codebase - type
/model, switch to Opus 5.5 for that part, then switch back. - Small background calls - send them to Claude Haiku 4.5.
My full setup and a cost breakdown are in Claude Code cost: pay per token, and rates for the whole Claude line in Claude Opus 5.5 API pricing.
Questions
Is Sonnet 5.5 a better pick than both?
It may be for many teams. Anthropic released Claude Sonnet 5.5 on September 28, and it sits in the catalog next to Sonnet 5 at $1.89 / $9.45 per 1M tokens. It has its own rules for thinking and tools, so test it before you move production. You can see every chat model with GET /v1/models?category=chat.
Why is my Opus bill higher than the rate gap suggests?
Because Opus 5.5 always reasons before it answers, and that reasoning is billed as output. Read the usage field on a few typical requests to see the real token counts.
Do both models share one key?
Yes. The same sk-kd- key and the same balance work for Opus, Sonnet and the rest of the catalog, so switching costs you nothing but a model name.