Model results
SWE-bench Pro V2
Choose a shared evaluation protocol to compare model scores. Versions, sources and configurations stay separate.
Atlas snapshot captured 10/19/2026, 10:20:28 AM UTC. This is an import timestamp, not the evaluation date. Coverage may be incomplete.
Reviewed recommendation and limits
Start with the locked public V2 protocol for contemporary repository issue repair. Choose Terminal-Bench for terminal workflows and LiveCodeBench for contest algorithms.
V2 public full split, 642 tasks across 11 repositories. Locked agent execution has a 50-minute task budget and fresh-container patch regrading. Preserve task revision, images, agent and settings. Imported groups below may use different protocols; this review does not certify each run. Reflection co-developed V2. Public tasks remain exposed and the full-set chart is near ceiling. Error-bar construction is not established in this review. Publisher pages disagree on correction counts. Private, HARD, V1 and generic Pro results do not establish this protocol.
No comparable scores imported yet
A methodology review does not mean results are imported. Use the publisher sources above for current results. We do not infer scores or ranks from a benchmark’s reputation.