DeepFinance benchmark: GPT-6 Astra max thinking leads at 57.6%

rahul @0interestratesOriginally posted 2026.11.08

Open original on X

after releasing DeepFinance bench, we wanted to see whether more thinking helps on ultra long-horizon finance tasks or if it just added “token-noise”. turns out, it helps. GPT-6 Astra on max thinking is the new leader on DeepFinance at 57.6%, exceeding Claude Opus 5.5 (54.8%) andClaude Sonnet 5.5 (54.6 %)

This post quotes another post: x.com/0interestrates/status/2118473925902635507