Claude Sonnet 5.5 narrows the gap with Anthropic’s top model

Anthropic has released Claude Sonnet 5.5, a mid-tier model that approaches its larger Opus sibling on several reported tests. On the GDPval-AA knowledge-work benchmark, Sonnet scored 1,844 points against Opus 5.5’s 1,846, according to Anthropic’s figures.

Coding showed the largest reported gain. Sonnet 5.5 reached 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5, and scored 55.5% on CursorBench 4.0, two points below Opus. At the maximum reasoning setting it performed worse than at the next level on one benchmark because extra sub-agents sometimes timed out or changed unrelated code.

Input and output prices remain $2 and $10 per million tokens, with cache reads at $0.20. Anthropic says lower token use cuts effective task cost by up to 30% and output is over 30% faster. Independent testing still needs to confirm those claims. The model is available through Anthropic and major cloud platforms with adjustable effort settings.