Anthropic released Claude Sonnet 5.5 on September 28, 2026, positioning it as the second model in the Claude 5.5 family after Opus 5.5 launched on September 22. The model keeps Sonnet 5's pricing at $2 per million input tokens and $10 per million output tokens, but Anthropic says it runs more than 30% faster and costs up to 30% less per task for most work.
Benchmark scores show a jump over Sonnet 5
Anthropic reported that Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0, a sharp rise from Sonnet 5's 10.3% and just above Opus 5.5's 66.4%. Independent benchmarking firm Artificial Analysis recorded a 64% score for Sonnet 5.5 on the same benchmark, compared to 60% for both Opus 5.5 and GPT-6 Astra. On Artificial Analysis's Intelligence Index, Sonnet 5.5 scored 56, placing it second overall, two points behind Opus 5.5 at maximum effort.
The independent firm also found that Sonnet 5.5 used roughly 193,000 output tokens per Intelligence Index task at maximum effort, about 60% more than Opus 5.5 and Sonnet 5 at max effort, and around seven times GPT-6 Astra's maximum-effort count. On knowledge-work tests such as GDPval-AA, Artificial Analysis found Sonnet 5.5 roughly level with Opus 5.5.
Finance tasks and token efficiency
Early-access customer Balyasny Asset Management reported that on its private suite of 2,441 finance tasks, Sonnet 5.5 scored ahead of Sonnet 5 and used about 121,000 tokens per answer, compared to Sonnet 5's 497,000. The reduction in token use suggests potential cost savings for firms running large volumes of queries, though the absolute token count remains higher than some competing models on certain benchmarks.
Safety and alignment focus
Anthropic stated that Sonnet 5.5 does not extend its capability frontier, so its alignment review focused on risks such as misleading users and aiding high-stakes misuse. The model ships with cyber safeguards and anti-distillation classifiers designed to block attempts to extract the model's reasoning for training rival systems. Those classifiers aim to protect Anthropic's intellectual property while allowing legitimate use.
Benchmark caveats and upcoming releases
Artificial Analysis tested a pre-release build of Sonnet 5.5 that carried a structured-output bug, which Anthropic has since fixed. The firm plans to rerun relevant evaluations, meaning the final performance numbers could shift slightly. Anthropic said Claude Haiku 5.5 is expected to follow in the coming weeks.
The launch lands in a busy month for AI model releases. On September 22, 2026, OpenAI released GPT-6 Sol and Luna the same day Anthropic launched Opus 5.5. OpenAI has shelved an upcoming model, GPT-6.1 Astra, on safety grounds. CNBC confirmed on Monday that the model fell short of the company's standards. The competitive pressure is clear, and Anthropic's decision to keep Sonnet 5.5's pricing flat while improving speed and cost-per-task may be a direct response to that landscape. What remains unresolved is how the fixed structured-output bug affects the model's benchmark standing once Artificial Analysis completes its rerun, and whether Haiku 5.5 will arrive before the end of the quarter.
