Claude Sonnet 5.5

Claude Sonnet 5.5 is the second model in Anthropic’s Claude 5.5 family, announced on 28 September 2026 as the successor to Sonnet 5 and served as claude-sonnet-5-5 on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry; TechCrunch and the AWS machine-learning blog reported it the same day. Anthropic positions it as the faster, cheaper choice for everyday work, coding and document creation, and says it costs up to 30 per cent less than Sonnet 5 for most work and runs more than 30 per cent faster, because it finishes tasks in fewer steps; the price per token itself is the list price below. The models overview gives a 1,000,000-token context window, 128,000 output tokens on the synchronous endpoint, a June 2026 reliable-knowledge cutoff, adaptive thinking with a default effort of high, and text and image input. The price is $2 per million input tokens and $10 per million output tokens, half the rate of Claude Opus 5.5, with cache reads at $0.20 and batch requests at half price. For planning, the useful line is the lifecycle: Anthropic’s table gives a retirement floor of not sooner than 28 September 2027 on Anthropic-operated platforms; Bedrock and Google Cloud set their own dates. Artificial Analysis, measuring independently, scores it 56 on version 4.3.2 of its Intelligence Index at maximum effort, against 58 for Opus 5.5, and records 410 million output tokens to complete the index and a long wait for the first token at that effort, so cost per task depends heavily on the effort setting chosen. Anthropic’s own figures — 70.6 per cent on Terminal-Bench 4.0, 46.2 per cent on FrontierCode 1.1, 55.5 per cent on CursorBench 4.0, 80.1 per cent on OSWorld 2.1 and 64.5 per cent on Humanity’s Last Exam with tools — are its own evaluation and are not stored as scores. Anthropic says a Haiku 5.5 model will follow in the coming weeks.