Inference
EN
Nebius

SWE-Rebench

A continuously updated software engineering benchmark by Nebius using fresh GitHub issues to avoid contamination. Models are evaluated 5 times per problem under a fixed ReAct scaffolding; the Resolved Rate (best pass@1) is reported.

ModelsModels with a result.13
Top resultBest result on this benchmark.—
Top-3 spreadPoints from first to third.—
YearYear of release.2026

Result and price

12 of 13 results

Results

13 results
#ModelResultThe score from the source.
—Claude Opus 4.665.3%
—GLM-562.8%
—GLM-5.162.7%
—DeepSeek V3.260.9%
—Claude Sonnet 4.660.7%
—Qwen3.5-27B58.9%
—GLM-4.758.7%
—Kimi K2.558.5%
—GPT-5.3 Codex58.2%
—Composer 258.0%
—Qwen3.5-35B-A3B53.7%
—MiniMax M2.751.9%
—Gemma 4 31B41.6%