Comparison · Chat & code models
Claude Opus 5.5 vs Claude Sonnet 5
Opus 5.5 or Sonnet 5? Opus costs exactly 2x more per token with the same context window. When Sonnet is enough in Claude Code and when Opus pays off.
Specs side by side
| 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 |
Price per token
| Claude Opus 5.5 | Claude Sonnet 5 | |
|---|---|---|
| Input / 1M tokens | $5.40 | $3.24 |
| Output / 1M tokens | $27.00 | $16.20 |
| Cached input / 1M | input rate | input rate |
| 1,000 replies (1,500 in + 400 out) | $18.90 | $11.34 |
Prices exclude taxes and are charged from your prepaid USD balance. Both models work through the same key, in the OpenAI-compatible /v1/chat/completions endpoint. Failed generations cost nothing.
For everyday coding Claude Sonnet 5 is enough: it is exactly half the price of Opus 5.5, with the same 1M-token window and the same 128K output limit. Bring in Claude Opus 5.5 for migrations across hundreds of files, reading a whole repository in one request, and agents that run unattended for hours. Both work in Claude Code through Jantaris, so you can switch between them mid-session with /model.
Key differences
- Price. Claude Opus 5.5 costs $5.40 / $27.00 per 1M tokens, Claude Sonnet 5 costs $3.24 / $16.20 per 1M tokens. Opus is 2 times more on input and output, and neither has a separate cached-input rate on Jantaris.
- Tier. Anthropic puts Opus 5.5 at Fable 5.1 level on most work, and Sonnet 5 close to Opus 4.8. That is one class apart, not a generation.
- Thinking. On Opus 5.5 adaptive thinking is always on and cannot be disabled. On Sonnet 5 it is on by default. In both, thinking tokens are billed as output.
- Parameters. Sonnet 5 rejects non-default
temperature,top_pandtop_kwith a 400 error. Remove them when porting older code. - Release. Sonnet 5 is from June 2026, Opus 5.5 from September 22, 2026 (knowledge cutoff June 2026).
- Shared. 1,000,000-token context, up to 128K output tokens, both endpoints:
/v1/chat/completionsand/v1/messages.
When to pick Claude Opus 5.5
- Refactors and migrations, say Flask to FastAPI, that touch hundreds of files.
- Claude Code agents that work on their own for many hours.
- Reading a whole repository, spec or contract set in one request.
- Planning: Opus breaks the task into steps and the cheaper Sonnet carries them out.
When to pick Claude Sonnet 5
- The default model in Claude Code: tests, fixes, new features, code review.
- Debugging legacy code down to the root cause.
- Agents with a few tools that run many times a day.
- Documentation, reports and emails at volume.
My Claude Code setup is Sonnet as ANTHROPIC_MODEL, a switch to Opus with /model for the hard part of a task, then back. Small background calls go to Claude Haiku 4.5. The full setup is in Claude Code cost: pay per token, and every Claude model's pricing is in Claude Opus 5.5 API pricing.
FAQ
Opus or Sonnet for coding?
Sonnet 5 for most work. Keep Opus 5.5 for tasks where one mistake costs more than double the rate: big migrations, long unattended sessions, whole-repository analysis.
How much more does Opus 5.5 cost than Sonnet 5?
Exactly 2 times on every input and output token: $5.40 / $27.00 per 1M tokens versus $3.24 / $16.20 per 1M tokens. Because Opus always thinks, the real per-request gap on short questions can be larger, so check the usage field.
How do I switch between Opus and Sonnet in Claude Code?
With /model during a session, or with the ANTHROPIC_MODEL variable before you start. Both models run on the same sk-jt- key and balance.
Will Sonnet 5.5 replace Sonnet 5?
Anthropic released Sonnet 5.5 on September 28 at the same price. The Jantaris catalog serves claude-sonnet-5 for now; the live list is at GET /v1/models?category=chat.
Does prompt caching lower the Opus bill?
On Jantaris, Opus 5.5 and Sonnet 5 have no separate cached-input rate, so a repeated prefix is billed as input (Opus 4.8 and Haiku 4.5 do have one). I go into more detail in Claude Opus 5.5 API pricing.