Anthropic released Claude Sonnet 5.5 on Monday, Sept. 28, the second model in its Claude 5.5 family and a faster, cheaper option for everyday AI tasks. The model runs 30 percent faster than its predecessor and costs up to 30 percent less to use, according to the company.
Sonnet 5.5 is designed for well-scoped work like code debugging, document creation, slides and spreadsheets. It sits between the budget-friendly Claude Haiku 5.5 (arriving soon) and the more powerful Claude Opus 5.5, which launched a week earlier.
Claude Sonnet 5.5 pricing and availability
Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, unchanged from Claude Sonnet 5, according to Reuters. The new model achieves lower per-task costs by requiring fewer tokens to complete the same work, according to Anthropic.
The model is available now on all of Anthropic’s platforms, per PC Gamer. Developers can access it through the API on Amazon Web Services, Google Cloud and Microsoft Azure, with zero data retention, according to The Verge. Users running Sonnet with thinking disabled must switch to a new between_tools setting before migrating.
What Sonnet 5.5 is designed for
The upgraded model includes enhanced cybersecurity safeguards that block attempts to breach its sandbox environment, according to The Verge. It’s the first Sonnet model to ship with cyber protections matching Opus 5.5’s approach, according to Mashable. Higher-risk cybersecurity requests will fall back to Sonnet 5, while routine bug-fixing remains unaffected.
Sonnet 5.5 also blocks distillation attacks, where bad actors extract a model’s capabilities at scale. Biology safeguards remain unchanged from Sonnet 5, though organizations seeking broader access can apply through Anthropic’s Life Sciences Verification Program, per Mashable.
Announced capabilities versus verified performance
Anthropic released benchmark results showing gains over Sonnet 5. On GDPval-AA v2.1, a test measuring real-world work across 44 occupations, Sonnet 5.5 scored 1,844, about 400 points ahead of Sonnet 5. It’s also the first Sonnet model to beat Pokémon Red using only screenshots. However, Anthropic cautioned that benchmarks capture only part of a model’s abilities, and outside testing found Opus 5.5 still clearly stronger at complex, open-ended work.
In early testing, Slack’s principal engineer Curtis Allen said Sonnet 5.5 beat Sonnet 5 on nearly all of Slack’s offline evaluations, finishing in fewer steps and with about 14 percent fewer output tokens. On Anthropic’s behavioral audit of roughly 1,850 scenarios, Sonnet 5.5 matched or improved on Sonnet 5 on most alignment measures, though Opus 5.5 still performed slightly better overall.
What’s next
Claude Haiku 5.5, designed for fast, high-volume work, is due in the coming weeks, according to Reuters. Anthropic is preparing for an IPO and has said enterprise customers, including Salesforce, Databricks, Goldman Sachs and Novo Nordisk, account for about 80 percent of its business.
Reporting from Reuters, The Verge and Mashable contributed to this story.
