Anthropic’s new Claude Sonnet 4.6 promises Opus-level coding at Sonnet pricing

Anthropic launches Claude Sonnet 4.6

Anthropic has released Claude Sonnet 4.6, the latest version of its mainstream AI model. The company says it nearly matches the performance of its flagship Opus 4.6 model, which launched two weeks ago, but at a significantly lower price.

Sonnet 4.6 is priced at $3 per million input tokens and $15 per million output tokens. This undercuts Opus 4.6, which costs $5 and $25 for the same token volumes. Like its predecessor, the new model offers a 1-million-token context window in beta.

Benchmark performance and improvements

On standard benchmarks, Sonnet 4.6 easily outperforms the previous Sonnet 3.5 model. It shows particularly strong results in office productivity tasks, even beating top competitors.

On OpenAI's GDPval benchmark for office work, Sonnet 4.6 outperforms Anthropic's own Opus 4.6, Google's Gemini 3 Pro, and OpenAI's GPT-5.2. For coding tasks, a traditional Anthropic strength, its results are often within a percentage point of Opus 4.6.

Key internal findings from Anthropic include:

• Developers preferred Sonnet 4.6 over the last-generation Opus 4.5 model 59% of the time.
• The new model is less likely to overengineer solutions.
• It shows a reduced tendency to hallucinate or falsely claim success.

Computer use capabilities

The model demonstrates advanced computer use skills, getting very close to Opus 4.6 on the OSWorld benchmark. Anthropic claims users are seeing "human-level capability" in complex, multi-step tasks.

These tasks include navigating intricate spreadsheets and filling out web forms across multiple browser tabs. However, Anthropic acknowledges that while models are approaching the human baseline on benchmarks, they still cannot match the most skilled human operators.

For developers using the API, Sonnet 4.6 now supports two key features in beta:

• Context compaction for managing longer sessions.
• Adaptive thinking, which lets the model set its own token budget for complex tasks based on their difficulty.

Availability and model lineup

Sonnet 4.6 will become the default model on claude.ai for users on both free and Pro plans. It will also be the default for Claude's Cowork mode, a logical fit given its strong performance on office-related tasks.

This update leaves Haiku, Anthropic's smallest and fastest model, as the only one not yet updated to a 4.6 release. Haiku 4.5 launched in October 2025, and Anthropic has historically updated this model at a slower pace, sometimes skipping version numbers.