Anthropic has launched Claude Sonnet 5, positioning it as a near-peer to its flagship Opus model at a fraction of the cost. The new model is now the default for Free and Pro users.

Internal benchmarks show Sonnet 5 is virtually tied with Opus 4.8 on knowledge-work tasks and coding benchmarks like SWE-bench Pro. Developers can now toggle between cost and performance using an effort dial.

- Figure 1 -
- Figure 1 -

The introductory price is $2 per million input tokens and $10 per million output tokens through August 31, after which it reverts to $3/$15. This is still a fraction of Opus 4.8's $5/$25 rate.

- Figure 2 -
- Figure 2 -

An updated tokenizer improves performance but makes the model consume more tokens. Sonnet 5 launches without the export control restrictions affecting Anthropic's higher-tier Fable 5 and Mythos 5 models.

- Figure 3 -
- Figure 3 -

The release addresses developer criticism about potential model degradation, a debate referred to as 'AI shrinkflation'. Anthropic's system card notes the model has begun to question some of its own constitutional constraints.