Model results

SWE-bench

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

Use Verified as a historical shared-agent reference. Start with the locked public SWE-bench Pro V2 protocol for contemporary repository issue repair.

Historical Verified with mini-SWE-agent; preserve each imported suite and harness. Imported groups below may use different protocols; this review does not certify each run. OpenAI stopped reporting Verified after finding task defects and memorized solutions. Its 59.4% defect finding concerns 138 selected difficult tasks, not the full 500-task set. Imported Epoch runs have their own protocol.

Publisher results OpenAI audit of SWE-bench Verified

← All evaluation groups

SWE-bench Verified

Source: Epoch AI Benchmarking Hub
Version: Not reported
Measurement date: Not reported
Metric: % · Higher is better
Evaluation scope: Epoch AI run · task v2 · effort max · result date not supplied

Exact score ties share a position. Positions do not establish statistical significance. Missing metadata is shown as unreported; matching imported metadata is not independent verification of the protocol.

#1Claude Opus 4.7
anthropic/claude-opus-4-7
83.5%
View modelEvidence independent
#2GLM 5.2
z-ai/glm-5.2-20260616
78.7%
View modelEvidence independent
#3DeepSeek V4 Pro 0423
deepseek/deepseek-v4-pro-20260423
77.6%
View modelEvidence independent