⌕Is ___ SOTA?
benchmarks / swe-bench-pro-hard

SWE-bench Pro V2 HARD

pass@1. Status active. Contamination risk med. Maintainer Scale Labs. Source.

ModelScoreDateKind
Claude Opus 5982026-10-12snapshot
Claude Fable 5.192.22026-10-12snapshot
GPT-6 Astra90.22026-10-12snapshot
Kimi K388.22026-10-12snapshot
Claude Sonnet 588.22026-10-12snapshot
GPT-5.6 Terra86.32026-10-12snapshot
GLM 5.384.32026-10-12snapshot
GPT-5.6 Sol82.42026-10-12snapshot
Claude Sonnet 5.570.592026-10-12snapshot
Gemini 3.8 Flash58.82026-10-12snapshot
Claude Haiku 5.552.942026-10-12snapshot
Claude Haiku 4.525.52026-10-12snapshot