Claude Sonnet 5.5

Claude Sonnet 5.5 is Anthropic's fast mid-tier AI model for everyday coding and knowledge work — near Opus 5.5 on many benchmarks at half the price.

What Changed

Sonnet 5.5, released September 28, 2026, is the second model in Anthropic's Claude 5.5 family, after Opus 5.5. It replaces Sonnet 5 from June: same per-token price, but it generates output over 30% faster and typically needs far fewer tokens per task, which Anthropic says makes it up to 30% cheaper to run. Anthropic positions it as the faster, lower-cost partner to Opus 5.5 — Opus for complex work that needs sustained judgment, Sonnet for well-scoped everyday tasks, bug fixes, and polished documents, slides, and spreadsheets. Claude Haiku 5.5 is due in the coming weeks.

How It Compares (Anthropic-Reported)

• Terminal-Bench 4.0 (agentic coding in a command line): 70.6%, versus 10.3% for Sonnet 5 and 66.4% for Opus 5.5 at its highest effort setting.
• CursorBench 4.0 (tasks from real Cursor coding sessions): 55.5%, versus 34.1% for Sonnet 5 and 57.8% for Opus 5.5.
• GDPval-AA (real-world tasks across 44 occupations): 1844 versus 1846 for Opus 5.5 and 1449 for Sonnet 5.
• OSWorld 2.1 (computer use): 80.1%, versus 57.0% for Sonnet 5 and 81.8% for Opus 5.5.
• FrontierCode 1.1: at High effort it matches GPT-6 Sol's best score at about a fifth of the cost per task, according to Anthropic.

These are Anthropic's own numbers. Anthropic itself says Opus 5.5 is still clearly stronger at complex, open-ended work. Independent tester Artificial Analysis also found that at max effort Sonnet 5.5 used more tokens per task than any model it had measured, so the cost savings depend on running it at Medium or High effort (the defaults: Medium in the Claude apps and Claude Code, High on the Claude Platform).

What Real Teams Are Saying

Anthropic published named quotes from teams that tested it on real work:

• Daniel Vogel, Chief Operating Officer at Epic Games — said it cleared the quality bar of a higher-tier model on a system design audit and a data flow review, and managed tens of thousands of lines of gameplay system architecture code.
• Abhinay Kathuria, Director of AI at Zendesk — fed it hundreds of real support cases; it made fewer wrong decisions and processed tickets 20% faster than the Claude models they use in production today.
• Fabian Hedin, Co-founder and CTO at Lovable — coding evals showed a third fewer tool calls and roughly half the shell runs to finish a task.
• Curtis Allen, Principal Engineer at Slack — without changing any prompts it beat Sonnet 5 on almost all of Slackbot's offline evals, in fewer steps and with about 14% fewer output tokens.

Pricing and Access

Input tokens cost $2 per million and output tokens $10 per million — the same as Sonnet 5 and half of Opus 5.5. Cache reads cost $0.20 per million, and the Batch API takes 50% off. It has a 1M-token context window and up to 128K tokens of output. The model ID is claude-sonnet-5-5 on the Claude API, Google Cloud, and Microsoft Foundry (anthropic.claude-sonnet-5-5 on Amazon Bedrock).

Migration Notes

• If you run Sonnet with thinking off, you need to switch to the new between_tools setting before moving to Sonnet 5.5.
• It's the first Sonnet model to ship with cyber safeguards: routine bug finding and fixing still works, but higher-risk cybersecurity requests visibly fall back to Sonnet 5.
• Images use a higher-resolution tier (up to 2576 pixels on the long edge), so image-heavy requests cost noticeably more tokens; downscale first if you don't need the detail.