What's new in Claude Sonnet 5
Simon Willison · Simon Willison · 2026-06-30
Simon Willison breaks down Claude Sonnet 5's release, finding that a new tokenizer producing ~30-40% more tokens for English text represents an effective hidden price increase despite unchanged nominal per-token rates.
Appears in
Extraction
Topics: anthropic-claudellm-releasellm-pricingtokenization
Claims
- Anthropic claims Sonnet 5 performance is close to Opus 4.8 while priced the same as Sonnet 4.6.
- Sonnet 5's new tokenizer produces approximately 30% more tokens than Sonnet 4.6 for the same input, making English text effectively ~1.4x more expensive.
- Sampling parameters temperature, top_p, and top_k are no longer supported in Sonnet 5.
- Sonnet 5 has a 1 million token context window and 128,000 maximum output tokens.
- Sonnet 5's reduced cyber-task capability relative to Mythos 5 allowed Anthropic to release it without US government blocking.
- Adaptive thinking is enabled by default in Sonnet 5 and must be explicitly disabled.
Key quotes
its performance is close to that of Opus 4.8, but at lower prices
The same input text produces approximately 30% more tokens than on Claude Sonnet 4.6.
Sonnet 5 is significantly less capable at cyber tasks than Mythos 5: its safeguards are thus similar to those we apply to Opus 4.7 and Opus 4.8 (models that are more capable than Sonnet 5 but much less capable than Mythos 5).