On September 28, 2026, Anthropic launched Claude Sonnet 5.5, the second model in the Claude 5.5 family after Opus 5.5. According to the company it is a clear upgrade over Claude Sonnet 5, runs more than 30% faster and costs up to 30% less for most work. The per-token price does not change: the savings come from the model consuming fewer tokens and making fewer tool calls.
Pricing and positioning
The API costs 2 dollars per million input tokens and 10 per million output tokens, the same as Sonnet 5. Anthropic presents it as the faster, cheaper complement to Opus 5.5: the latter remains the model for complex work that needs careful judgment, while Sonnet 5.5 is best for well-scoped everyday tasks such as fixing bugs or creating polished documents, slides and spreadsheets.
The stated benchmarks
Anthropic compares Sonnet 5.5 with Sonnet 5 and, in some cases, with Opus 5.5:
| Test | Sonnet 5.5 | Sonnet 5 | Opus 5.5 |
|---|---|---|---|
| CursorBench 4.0 (agentic coding) | 55.5% | 34.1% | – |
| OSWorld 2.1 (computer use) | 80.1% | 57.0% | – |
| GDPval-AA v2.1 (professional work, Elo score) | 1,844 | 1,449 | 1,846 |
| Humanity's Last Exam (with tools) | 64.5% | 54.9% | – |
The most interesting figure is office-type work: on GDPval-AA, Sonnet 5.5 is practically level with Opus 5.5 (1,844 against 1,846) while being the model designed to be faster and cheaper. These are company-reported results: as always, check them against your own tasks.
What early customers say
Anthropic reports a few testimonials. Epic Games says the model handled tens of thousands of lines of code for a game-system architecture and multi-hour tasks with less prescriptive prompting. CodeRabbit notes that two Sonnet 5 weaknesses, excessive reliance on web search and high token usage, have disappeared.
For documents, Anthropic cites a 10-slide operations review that two experts judged ready to send as it was.
Safety updates
Sonnet 5.5 is the first Sonnet model with cyber safeguards similar to Opus 5.5's: high-risk cybersecurity tasks will be blocked more often than with Sonnet 5, while routine software development stays allowed. For legitimate security needs there is a dedicated verification program.
It is also the first Sonnet with classifiers designed to prevent extraction of the model's reasoning, the technique known as distillation. On about 1,850 automated behavioral-audit scenarios, Anthropic says Sonnet 5.5 improves on or matches Sonnet 5 on most measures of alignment, resistance to misuse and honesty.
Where to use it
- Claude.ai, immediately
- Claude Platform API, with the identifier
claude-sonnet-5-5 - Amazon Web Services, Google Cloud and Microsoft Azure
- With a zero data retention option on all platforms
Claude Haiku 5.5, the version for high-volume, cost-sensitive applications, will arrive "in the coming weeks".
Why it matters
The comparison between top-tier models is shifting from raw power to cost per task. Sonnet 5.5 bets on that: same per-token price, but fewer tokens and fewer steps to reach the same result. For anyone building tools or automations, that is the number to watch, along with quality on your own use cases.
Read also: OpenAI launches GPT-6.1 Sol and Google unveils Gemini 4 Argon.