Anthropic's first hybrid reasoning model, letting users toggle standard and extended-thinking modes. Anthropic publishes no single-attempt extended-thinking GPQA figure for this model, so the standard-mode score is recorded.

Benchmarks

What its maker published, and what anyone else measured. Every figure links the document it came from.

benchmark	published	measured
GPQA	68	—
SWE-bench	63.7	52.8

Lineage

No recorded lineage — root or standalone model.

view in full graph →

Metadata

org	Anthropic
released	2025-02-24
license	proprietary