What's new in Claude Sonnet 5

Anthropic has released Claude Sonnet 5, a new AI model that offers performance comparable to the previous Opus 4.8 model at a lower price point. The model features a 1 million token context window and a 128,000 maximum output token limit. It retains the same toolset and platform features as the previous Sonnet 4.6 version, with adaptive thinking enabled by default.

A significant change in this release is the implementation of a new tokenizer. While the base pricing remains the same as the previous version, the new tokenizer causes the same input text to generate approximately 30 percent more tokens. This effectively results in a price increase of roughly 28 to 42 percent for English, Spanish, and Python code, depending on the content, while costs for Simplified Mandarin remain largely unchanged.

From a technical standpoint, the model no longer supports the sampling parameters of temperature, top_p, and top_k. Anthropic notes that the model is less capable at cyber tasks than the Mythos 5 model, which allowed for a smoother regulatory approval process. These updates provide developers with a more efficient, albeit more expensive, tool for large-scale language processing tasks.