Claude Sonnet 5 matches Opus on benchmarks — and costs far less
anthropic released claude sonnet 5, which beats its predecessor sonnet 4.6 and rivals the more expensive opus 4.8 on benchmarks like coding and knowledge work. it signals the 'smaller, smarter' trend is accelerating, offering near-frontier performance at a lower price.
anthropic released claude sonnet 5, marketing it as the most agentic version of the model yet with improved planning and tool use for browsers and terminals. benchmark results indicate a clear jump over sonnet 4.6, with sonnet 5 scoring 63.2% on swe-bench pro (coding) and 81.2% on osworld-verified (computer use), closing the gap with opus 4.8. notably, it scored 1,618 on the gdpval-aa v2 knowledge work benchmark, slightly surpassing opus 4.8's 1,615. the launch follows us government blocks on mythos 5 and fable 5, so anthropic says sonnet 5 is not trained on cyber tasks and features new safeguards to mitigate exploit risks.