Anthropic releases Claude Sonnet 5 with 1M context window and new tokenizer

Anthropic has released Claude Sonnet 5, a new model whose performance is described as close to Opus 4.8 but at lower prices. The release includes significant API changes, most notably a new tokenizer that produces approximately 30% more tokens than Claude Sonnet 4.6 for the same input.

• The model features a 1 million token context window and supports up to 128,000 output tokens.
• Sampling parameters temperature, top_p, and top_k are no longer supported in the API.
• Adaptive thinking is enabled by default unless explicitly disabled via the API.
• Pricing remains $3/million input and $15/million output tokens, though the new tokenizer effectively raises costs by roughly 28-42% depending on language.
• The model retains the same set of tools and platform features as Claude Sonnet 4.6.

The new tokenizer significantly increases token counts for English, Spanish, and Python code compared to previous versions, while Simplified Mandarin tokenization remains largely unchanged.