Claude Sonnet 5 Released: New Tokenizer Boosts Efficiency but Costs 30% More
Decision Brief
What changedAnthropic releases Claude Sonnet 5 with performance near Opus 4.8 at lower price, but new tokenizer raises effective cost by ~30%.
Why it mattersAI builders must assess real cost impact of tokenizer changes and API parameter removals (temperature, top_p, top_k).
Who should careTeams building on model APIs
Affected stackClaude
Builder actionEvaluate
Source confidenceMedium · Reliable media or first-hand reporting
Anthropic launched Claude Sonnet 5 today, claiming performance close to Opus 4.8 at a lower price. Major API changes: no longer supports temperature, top_p, top_k sampling parameters; context window of 1 million tokens with max output of 128k tokens; adaptive thinking enabled by default. Pricing matches Sonnet 4.6 ($3/$15 per million tokens), with a discount until Aug 31 ($2/$10). However, the new tokenizer increases token count for the same input text by ~30%. Tests show English docs increase 1.42x, Spanish 1.33x, Python code 1.28x, while Simplified Chinese remains roughly unchanged. Developers need to reevaluate actual usage costs.
Summary basis: official / RSS sourceUnless it says 'full article read', this summary is based only on publicly available content — it never pretends to have read restricted originals.
Sources
- Simon Willison：Blog
Hands-on notes on LLM tools, local models, and practical AI engineering.
- Simon Willison：Blog