Model results
SWE-rebench
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.
SWE-rebench
- Source
- SWE-rebench (reviewed publisher results)
- Version
- 2026-05-15_2026-07-01
- Measurement date
- Not reported
- Metric
- % · Higher is better
- Evaluation scope
- pass-at-5; 111 problems from 65 repositories; task window 2026-05-15 to 2026-07-01; model track; configuration DeepSeek-V4 Pro [high]; tools interaction mode; minimal ReAct framework; pass@5 across five full-benchmark runs; published resolved mean 40.2% with SEM 1.29 percentage points; SEM is standard error of the mean, not a confidence interval or standard deviation; publisher context policy 128K tokens unless the model supports less; developer-recommended default generation settings; declared-model match to Atlas deepseek/deepseek-v4-pro-20260423; exact execution backend and API request model identifier unverified; publisher model release date 2026-04-24 is not a measurement date or proof of the Atlas dated execution; original V4 Pro declaration with high effort, not the August 13 successor; publisher does not flag potential contamination for this window; this does not prove decontamination; result date not supplied; source captured 2026-10-19
Fewer than two comparable models. No ranking is assigned. Missing metadata is shown as unreported; matching imported metadata is not independent verification of the protocol.
DeepSeek V4 Pro 0423
deepseek/deepseek-v4-pro-20260423
64%
View modelEvidence independent