What's new in Claude Sonnet 5

Simon Willison analyzed the developer docs for Claude Sonnet 5, released today by Anthropic. According to Willison, Anthropic claims the model's performance is close to that of Opus 4.8 but at lower prices. However, a new tokenizer produces approximately 30% more tokens for the same input compared to Sonnet 4.6, effectively negating the cost advantage. The model's safety classification is lower than Mythos 5 for cyber tasks, so it faces similar regulatory safeguards as Opus 4.7 and 4.8. Notable API changes include the removal of sampling parameters (temperature, top_p, top_k) and the default enabling of adaptive thinking, which can be disabled. Sonnet 5 supports a 1 million token context window and 128,000 maximum output tokens. Pricing matches Sonnet 4.6 at $3/million input and $15/million output, with an introductory discount to $2/$10 until August 31. For developers integrating AI into workflows, the tokenizer change is a hidden cost increase, and the loss of sampling parameters reduces fine-grained control over output randomness. Willison highlighted these practical implications using his Token Counter tool.