Calibration
SWE-bench Verified → SWE-Rebench
SWE-Rebench is estimated from SWE-bench Verified with a offset logistic curve fitted on 8 models measured on both: y = 0.5703 + (0.6177 − 0.5703) / (1 + exp(−200.00·(x − 0.7686))), R² = 0.44, cross-validated error 3.6 pp. It is used for 82 estimates.
| Estimated model | SWE-bench Verified | SWE-Rebench | Source |
|---|---|---|---|
| Claude Fable 5 | 95.0% | 61.8% | estimated ± 3.6 pp, low confidence |
| Claude Mythos 5 | 95.5% | 61.8% | estimated ± 3.6 pp, low confidence |
| Claude Opus 5 | 96.0% | 61.8% | estimated ± 3.6 pp, low confidence |
| Ember-1 | 92.2% | 61.8% | estimated ± 3.6 pp, low confidence |
| Claude Opus 4.8 | 88.6% | 61.8% | estimated ± 3.6 pp, low confidence |
| Claude Opus 4.7 (Adaptive) | 87.6% | 61.8% | estimated ± 3.6 pp, low confidence |
| Ornith-1.5-397B | 86.0% | 61.8% | estimated ± 3.6 pp, low confidence |
| Claude Sonnet 5 | 85.2% | 61.8% | estimated ± 3.6 pp, low confidence |
| Ornith-1.0-397B | 82.4% | 61.8% | estimated ± 3.6 pp, medium confidence |
| Beam 1.0 501B-A23B | 81.6% | 61.8% | estimated ± 3.6 pp, medium confidence |
| Claude Opus 4.5 | 80.9% | 61.8% | estimated ± 3.6 pp, medium confidence |
| Beam 1.0 501B-A23B NVFP4 | 80.7% | 61.8% | estimated ± 3.6 pp, medium confidence |
| DeepSeek V4 Pro 0813 | 80.6% | 61.8% | estimated ± 3.6 pp, medium confidence |
| MiniMax M3 | 80.5% | 61.8% | estimated ± 3.6 pp, medium confidence |
| Qwen3.7 Max | 80.4% | 61.8% | estimated ± 3.6 pp, medium confidence |
| Inkling-Small | 80.2% | 61.8% | estimated ± 3.6 pp, medium confidence |
| Kimi K2.6 | 80.2% | 61.8% | estimated ± 3.6 pp, medium confidence |
| Beam 1.0 501B-A23B FP8 | 80.2% | 61.8% | estimated ± 3.6 pp, medium confidence |
| GPT-5.2 | 80.0% | 61.8% | estimated ± 3.6 pp, medium confidence |
| DeepSeek V4 Flash 0731 | 79.0% | 61.7% | estimated ± 3.6 pp, medium confidence |
| Ornith-1.5-35B-A3B | 79.0% | 61.7% | estimated ± 3.6 pp, medium confidence |
| Qwen3.6 Plus | 78.8% | 61.7% | estimated ± 3.6 pp, medium confidence |
| BTL-4 | 78.4% | 61.6% | estimated ± 3.6 pp, medium confidence |
| dots3-note Preview | 78.4% | 61.6% | estimated ± 3.6 pp, medium confidence |
| MiMo-V2-Pro | 78.0% | 61.3% | estimated ± 3.6 pp, medium confidence |
| Apodex 1.1 | 77.7% | 61.0% | estimated ± 3.6 pp, medium confidence |
| Qwen3.7 Plus | 77.7% | 61.0% | estimated ± 3.6 pp, medium confidence |
| Inkling | 77.6% | 60.9% | estimated ± 3.6 pp, medium confidence |
| Mistral Medium 3.5 128B | 77.6% | 60.9% | estimated ± 3.6 pp, medium confidence |
| Muse Spark | 77.4% | 60.6% | estimated ± 3.6 pp, medium confidence |
| Claude Sonnet 4.5 | 77.2% | 60.2% | estimated ± 3.6 pp, medium confidence |
| Qwen3.6-27B | 77.2% | 60.2% | estimated ± 3.6 pp, medium confidence |
| Kimi K2.5 (Reasoning) | 76.8% | 59.3% | estimated ± 3.6 pp, medium confidence |
| Grok 4.20 | 76.7% | 59.0% | estimated ± 3.6 pp, medium confidence |
| Qwen3.5 397B | 76.2% | 58.0% | estimated ± 3.6 pp, medium confidence |
| Muse Glimmer 30B | 76.0% | 57.8% | estimated ± 3.6 pp, medium confidence |
| Ornith-1.0-35B | 75.6% | 57.4% | estimated ± 3.6 pp, medium confidence |
| MiMo-V2-Omni | 74.8% | 57.1% | estimated ± 3.6 pp, medium confidence |
| Laguna M.1 | 74.6% | 57.1% | estimated ± 3.6 pp, medium confidence |
| Claude 4.1 Opus | 74.5% | 57.1% | estimated ± 3.6 pp, medium confidence |
| Hy3 Preview | 74.4% | 57.1% | estimated ± 3.6 pp, medium confidence |
| Step 3.5 Flash | 74.4% | 57.1% | estimated ± 3.6 pp, medium confidence |
| Flint | 74.2% | 57.1% | estimated ± 3.6 pp, medium confidence |
| MAI-Thinking-1 | 73.5% | 57.0% | estimated ± 3.6 pp, medium confidence |
| MiMo-V2-Flash | 73.4% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Qwen3.6-35B-A3B | 73.4% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Claude Haiku 4.5 | 73.3% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Jais-3-MoVA-36B-A4B | 72.8% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Claude 4 Sonnet | 72.7% | 57.0% | estimated ± 3.6 pp, medium confidence |
| MAI-Code-1.1-Flash | 72.6% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Claude 4 Opus | 72.5% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Jais-3-32B | 72.4% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Qwen3.5-122B-A10B | 72.0% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Nemotron 3 Ultra | 71.9% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B | 71.1% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B NVFP4 | 71.1% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Laguna XS 2.1 | 70.9% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Grok Code Fast 1 | 70.8% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Ornith-1.5-9B | 70.6% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Solar Pro 4 | 70.6% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Solar Open 2 | 70.4% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Jais-3-375B-A23B | 70.2% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Jais-3-7B | 70.2% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Laguna XS.2 | 69.9% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Qwen3-Coder 480B-A35B | 69.6% | 57.0% | estimated ± 3.6 pp, medium confidence |
| Ornith-1.0-9B | 69.4% | 57.0% | estimated ± 3.6 pp, medium confidence |
| K-EXAONE 2.0 | 68.2% | 57.0% | estimated ± 3.6 pp, low confidence |
| LongCat-Flash-Lite-Sparse | 68.2% | 57.0% | estimated ± 3.6 pp, low confidence |
| DeepSeek V3.1 | 66.0% | 57.0% | estimated ± 3.6 pp, low confidence |
| Kimi K2 | 65.8% | 57.0% | estimated ± 3.6 pp, low confidence |
| Gemini 2.5 Pro | 63.8% | 57.0% | estimated ± 3.6 pp, low confidence |
| Spark-X2.6-4B | 62.8% | 57.0% | estimated ± 3.6 pp, low confidence |
| GPT-OSS 120B | 62.4% | 57.0% | estimated ± 3.6 pp, low confidence |
| Claude 3.7 Sonnet | 62.3% | 57.0% | estimated ± 3.6 pp, low confidence |
| Ternary Bonsai 2 27B | 60.8% | 57.0% | estimated ± 3.6 pp, low confidence |
| GPT-OSS 20B | 60.7% | 57.0% | estimated ± 3.6 pp, low confidence |
| Nemotron 3 Super 100B | 60.5% | 57.0% | estimated ± 3.6 pp, low confidence |
| DeepSeek-R1-0528 | 57.6% | 57.0% | estimated ± 3.6 pp, low confidence |
| GLM-4.7-Flash | 59.2% | 57.0% | estimated ± 3.6 pp, low confidence |
| GPT-4.1 | 54.6% | 57.0% | estimated ± 3.6 pp, low confidence |
| Granite 4.2 30B | 57.0% | 57.0% | estimated ± 3.6 pp, low confidence |
| MiniMax M1 80k | 56.0% | 57.0% | estimated ± 3.6 pp, low confidence |
Fitted on
| Model | SWE-bench Verified | SWE-Rebench |
|---|---|---|
| Claude Opus 4.6 | 80.8% | 65.3% |
| Claude Sonnet 4.6 | 79.6% | 60.7% |
| GLM-4.7 | 73.8% | 58.7% |
| GLM-5 | 77.8% | 62.8% |
| GPT-5.3 Codex | 85.0% | 58.2% |
| Kimi K2.5 | 76.8% | 58.5% |
| Qwen3.5-27B | 72.4% | 58.9% |
| Qwen3.5-35B-A3B | 69.2% | 53.7% |