Anthropic launches Claude Sonnet 5.5 with new benchmark results

Anthropic has launched Claude Sonnet 5.5 on September 28, 2026, and says the model is available on all platforms. It is aimed at everyday, well-scoped tasks such as bug fixes and creating documents, slides, and spreadsheets; Anthropic positions Claude Opus 5.5 for more complex work requiring sustained judgment.

Anthropic says Sonnet 5.5 generates output more than 30% faster than Claude Sonnet 5 and can cost up to 30% less per task for most work, typically because it uses fewer tokens. The per-token API rates remain unchanged, though—and the total bill for a task depends on how many tokens it uses.

API rates and task costs

Anthropic lists these API rates in U.S. dollars per million tokens:

• Input: $2
• Output: $10
• Cache reads: $0.20
• Cache writes: $2.50

A token is a small unit of text the model processes or generates. These rates are charges for API usage, not a consumer subscription price.

A separate independent evaluation of a pre-release deployment offers a more specific cost comparison. At Max effort, it estimated $7.60 per Intelligence Index task, about 50% more than Claude Sonnet 5 in that evaluation. The same test used approximately 193,000 output tokens per task. That estimate applies to the evaluation’s workload and settings; it does not replace Anthropic’s claim about costs across most work.

The pre-release deployment had a structured-output bug. Anthropic said it fixed the issue for public release and expected its effect to be small or to understate performance.

What Anthropic’s benchmarks report

Anthropic’s published comparison gives Claude Sonnet 5.5 a 70.6% score on Terminal-Bench 4.0. The same table lists 10.3% for Claude Sonnet 5 and 66.4% for Claude Opus 5.5 at Xhigh effort. On GDPval-AA, which assesses work across 44 occupations, the scores are close for the two newer Claude models.

The separate Max-effort evaluation reported a score of 56 on its Intelligence Index. Its setup differs from Anthropic’s published benchmarks, so those figures describe different tests rather than a single head-to-head result.

A concrete output from Sonnet 5.5

Anthropic’s clock-of-clocks example gives a look at a visual output associated with the launch: circular clock faces form large digital-style numerals.

Cybersecurity safeguards and API settings

Anthropic says higher-risk cybersecurity tasks visibly fall back to Claude Sonnet 5, while routine software development is unaffected. It describes Sonnet 5.5’s biology safeguards as the same as Sonnet 5’s, and says Sonnet 5.5 is its first Sonnet model to launch with classifiers designed to prevent reasoning extraction in distillation attacks.

The API identifier is claude-sonnet-5-5. Anthropic sets the default effort to Medium in Claude apps and Claude Code, and to High on the Claude Platform. API users who run Sonnet with thinking turned off must switch to the between_tools setting.

The next model in the Claude 5.5 family

Anthropic says Claude Haiku 5.5 will join the family “in the coming weeks.”