Claude Sonnet 5.5 price, benchmarks and when to pick it
Claude Sonnet 5.5 price stays at $2 and $10 per million tokens. Which scores are Anthropic's own, what outside tests show, and when Opus 5.5 is the better buy.
Source-based. Written from the documents, reporting and reviews linked in the text. Nothing here was tested hands-on by The Ruling Desk. How we work
Anthropic released Claude Sonnet 5.5 on September 28, 2026, and the Claude Sonnet 5.5 price is the same as Sonnet 5's: $2 per million input tokens and $10 per million output tokens. Anthropic says it runs more than 30% faster than Sonnet 5 and scores 70.6% on Terminal-Bench 4.0, ahead of the pricier Opus 5.5. Most of those numbers are Anthropic's own, and the one independent index so far tells a more mixed story about cost.
Key takeaways
- Per-token prices are unchanged from Sonnet 5, and half of Opus 5.5's, according to Anthropic's pricing docs as of September 29, 2026. Cache reads cost $0.20 per million tokens on both Sonnet 5.5 and Opus 5.5.
- Anthropic's benchmark table has Sonnet 5.5 ahead of Opus 5.5 only on Terminal-Bench 4.0 (70.6% to 66.4%). Opus 5.5 leads on the other seven results listed for both.
- Independently, Artificial Analysis scores Sonnet 5.5 at 56 on its Intelligence Index at max effort, up from 38 for Sonnet 5, and measures 85 to 137 output tokens per second.
- The same index shows Sonnet 5.5 cheaper per task than Sonnet 5 at medium and high effort, but about 49% pricier at max effort, where Opus 5.5 scores higher for less.
- It's on the Claude API as
claude-sonnet-5-5, plus Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS.
What the Claude Sonnet 5.5 price means next to Sonnet 5 and Opus 5.5
Every line on Anthropic's pricing page is identical for Sonnet 5 and Sonnet 5.5. Prices are in US dollars per million tokens, as listed on September 29, 2026:
| Item | Sonnet 5 | Sonnet 5.5 | Opus 5.5 |
|---|---|---|---|
| Input | $2 | $2 | $4 |
| Output | $10 | $10 | $20 |
| Cache write (5 minutes) | $2.50 | $2.50 | $5 |
| Cache write (1 hour) | $4 | $4 | $8 |
| Cache read | $0.20 | $0.20 | $0.20 |
| Batch input / output | $1 / $5 | $1 / $5 | $2 / $10 |
The same page notes that Sonnet 5's $2 and $10 started as an introductory price through August 31, 2026, and that the planned rise to $3 and $15 on September 1 "will not occur." So Sonnet 5.5 locks in the lower rate. It's also the same list price as OpenAI's GPT-6 Sol.
In plain numbers, here's our arithmetic at list prices: a job that sends 10 million input tokens and gets back 2 million output tokens costs $40 on Sonnet 5.5 and $80 on Opus 5.5. The exception is cache reads, where Opus 5.5 got a steeper discount and now costs the same $0.20.
Anthropic's announcement claims a saving beyond the price list: "In our testing, it costs up to 30% less per task than its predecessor." That depends on the model using fewer tokens for the same job, which you can only confirm on your own bills.
Claude Sonnet 5.5 benchmarks: whose numbers are whose
The headline scores come from Anthropic's table. Its footnotes say Opus 5.5's Terminal-Bench result is at xhigh effort, and that Artificial Analysis ran GDPval-AA and AA-Briefcase on a pre-release deployment with a structured-output bug that has since been fixed. The page doesn't name an outside evaluator for the other rows.
| Benchmark (Anthropic's table) | Sonnet 5.5 | Sonnet 5 | Opus 5.5 |
|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 70.6% | 10.3% | 66.4% |
| FrontierCode 1.1 | 46.2% | 42.4% | 54.4% |
| CursorBench 4.0 | 55.5% | 34.1% | 57.8% |
| GDPval-AA v2.1 (run by Artificial Analysis, Elo) | 1844 | 1449 | 1846 |
| AA-Briefcase v1.1 (run by Artificial Analysis, Elo) | 1811 | 1359 | 1822 |
| Humanity's Last Exam, with tools | 64.5% | 54.9% | 67.7% |
| OSWorld 2.1 (computer use) | 80.1% | 57.0% | 81.8% |
| Chartography, no tools | 61.6% | 15.6% | 64.4% |
The jump from 10.3% to 70.6% on Terminal-Bench is the eye-catcher, and Anthropic's page doesn't explain why Sonnet 5 scored so low. Against Opus 5.5, the gaps run from 2 Elo points on GDPval-AA, effectively a tie, to 8.2 points on FrontierCode. Anthropic also notes that Sonnet 5.5 scored lower at max effort than at xhigh on FrontierCode, because the extra effort set off a code-review step that caused timeouts.
What independent testing shows so far
Artificial Analysis, which runs its own evaluations, scores Sonnet 5.5 at 56 on its Intelligence Index v4.3.2 at max effort and 47 at high effort, as of September 29, 2026. Its Sonnet 5 page shows 38 and 32 on the same index. That backs Anthropic's claim of a big step up.
The speed claim holds up too: Artificial Analysis measured 85 to 137 output tokens per second across effort levels, against 56 to 75 for Sonnet 5.
The cost-per-task claim is where it gets mixed. On Artificial Analysis' suite, Sonnet 5.5 costs $0.59 per task at medium effort and $1.08 at high, against $1.00 and $1.79 for Sonnet 5. At max effort it costs $7.60, about 49% more than Sonnet 5's $5.09, because higher effort burns far more tokens.
Sonnet 5.5 vs Opus 5.5: when to pick which
Anthropic's own pitch is that Opus 5.5 "is built for complex work requiring careful judgment," while Sonnet 5.5 is "strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets."
The independent numbers add a twist. On Artificial Analysis' index, Opus 5.5 scores 58 at max effort for $5.98 per task, higher and cheaper than Sonnet 5.5's 56 at $7.60. Opus 5.5 at medium effort scores 51 for $1.34, close to Sonnet 5.5 at xhigh (52 for $2.74). Our reading: for the hardest work, Opus 5.5 at a lower effort setting can be the better buy. That's one composite index, though, not your workload.
Sonnet 5.5 makes more sense when:
- Speed matters. It was faster than Opus 5.5 at every effort level Artificial Analysis measured.
- Your prompts are long and your answers short, like document review or retrieval, since input costs half as much as on Opus 5.5.
- You run at low to high effort on routine coding and office tasks, where it's cheap per task.
What changes if you switch from Sonnet 5
According to the model overview, Sonnet 5.5 has a 1 million token context window, up to 128,000 output tokens, and an Amazon Bedrock ID of anthropic.claude-sonnet-5-5. Anthropic commits to keeping it available until at least September 28, 2027.
The docs list five breaking changes for code built on Sonnet 5. Among them: you turn off up-front thinking with a new between_tools setting, and forcing a specific tool call returns an error. Separately, setting temperature, top_p or top_k to a non-default value is rejected. Default effort is high on the API, while Anthropic says Claude Code and its apps default to medium.
Anthropic says Haiku 5.5 will follow "in the coming weeks," with no date or price. For the company's business side, see what Anthropic's IPO prospectus shows. OpenAI's lineup is shifting too: it canceled GPT-6.1 Astra.
Bottom line
Claude Sonnet 5.5 costs the same per token as Sonnet 5, half of Opus 5.5, and Artificial Analysis confirms it's much smarter and faster than Sonnet 5. The Terminal-Bench win over Opus 5.5 is Anthropic's own number, and on the independent index Opus 5.5 still leads, cheaply at max effort. If you run Sonnet 5, switch after checking the breaking changes. If you need top quality, price Opus 5.5 at medium effort before defaulting to Sonnet. More coverage lives in our AI section.
FAQ
How much does Claude Sonnet 5.5 cost?
The Claude Sonnet 5.5 price on the Claude API is $2 per million input tokens and $10 per million output tokens, as of September 29, 2026. Cache reads cost $0.20 per million and the Batch API halves input and output prices. Bedrock and Google Cloud publish their own pricing.
Is Claude Sonnet 5.5 better than Opus 5.5?
Only on Terminal-Bench 4.0 in Anthropic's own table, 70.6% to 66.4%. Opus 5.5 leads on the other seven results listed for both, and on Artificial Analysis' independent index, 58 to 56 at max effort. Sonnet 5.5 is faster and half the price per token.
When will Claude Haiku 5.5 come out?
Anthropic only says "in the coming weeks." It hasn't announced a date or a price.