Claude Sonnet 5.5

Everyday work, coding and document creation are where Anthropic positions Claude Sonnet 5.5, the faster and cheaper choice in its lineup. Anthropic announced it on 28 September 2026 as the follow-up to Sonnet 5 and the second model of the Claude 5.5 generation. Its model ID is claude-sonnet-5-5, and it is offered through the Claude API as well as Microsoft Foundry, Google Cloud and Amazon Bedrock. Coverage came the same day from TechCrunch and the AWS machine-learning blog.

What Anthropic says changed

Anthropic says that for most work it costs up to 30 percent below Sonnet 5 and finishes more than 30 percent sooner, since tasks need fewer steps. The per-token rate is half that of Claude Opus 5.5, with discounts for cache reads and for batch requests.

Specifications

The models overview gives a context window of 1,000,000 tokens, a ceiling of 128,000 output tokens when called synchronously, a reliable-knowledge cutoff of June 2026, and text and image input. Adaptive thinking is built in, with a default effort of high.

Planning around the lifecycle

On platforms Anthropic runs itself, its table sets a retirement floor of 28 September 2027 at the earliest, while Bedrock and Google Cloud set dates of their own. That line matters for teams deciding how long they can build on the model.

Independent and vendor measurements

In independent testing, Artificial Analysis gives it 56 on the Intelligence Index (version 4.3.2) at maximum effort, with Opus 5.5 on 58. Completing the index took 410 million output tokens, and the wait for a first token at that effort was long, which means the cost of a task depends heavily on the effort setting. Anthropic's own figures are 70.6 percent on Terminal-Bench 4.0, 46.2 percent on FrontierCode 1.1, 55.5 percent on CursorBench 4.0, 80.1 percent on OSWorld 2.1 and 64.5 percent on Humanity's Last Exam with tools. A Haiku 5.5 model is promised by Anthropic within weeks.

Claude Sonnet 5.5 is a mid-tier model from Anthropic, the second in its Claude 5.5 family and the successor to Sonnet 5. It is aimed at everyday work, coding and document creation. Artificial Analysis scored it below Claude Opus 5.5 on its independent index, while Anthropic's own benchmark figures are not independent.

Main features

• Everyday tasks: positioned as the faster choice for routine work, coding and documents, finishing tasks in fewer steps.
• Large context: a 1,000,000-token context window and up to 128,000 output tokens on the synchronous endpoint.
• Input types: text and image input.
• Adaptive thinking: reasoning effort adjusts to the task, with a default setting of high.
• Access routes: the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry, as served at launch.

Who it serves

Developers and teams that call the model through the Claude API or a cloud platform for coding, documents and everyday work.

Key facts

• Developer: Anthropic.
• Released: 28 September 2026.
• Knowledge cutoff: June 2026, according to the models overview.
• Retirement: not sooner than 28 September 2027 on Anthropic-operated platforms, while cloud providers set their own dates.