Calibration
SWE-bench Verified → Vals SWE-bench
Vals SWE-bench is estimated from SWE-bench Verified with a linear curve fitted on 24 models measured on both: y = 1.3755·x + -0.3368, R² = 0.69, cross-validated error 6.8 pp. It is used for 48 estimates.
| Estimated model | SWE-bench Verified | Vals SWE-bench | Source |
|---|---|---|---|
| Claude Mythos 5 | 95.5% | 97.7% | estimated ± 6.8 pp, medium confidence |
| Ember-1 | 92.2% | 93.1% | estimated ± 6.8 pp, medium confidence |
| BTL-4 | 78.4% | 74.2% | estimated ± 6.8 pp, medium confidence |
| MiMo-V2-Pro | 78.0% | 73.6% | estimated ± 6.8 pp, medium confidence |
| Claude Sonnet 4.5 | 77.2% | 72.5% | estimated ± 6.8 pp, medium confidence |
| MiMo-V2-Omni | 74.8% | 69.2% | estimated ± 6.8 pp, medium confidence |
| Claude 4.1 Opus | 74.5% | 68.8% | estimated ± 6.8 pp, medium confidence |
| Step 3.5 Flash | 74.4% | 68.7% | estimated ± 6.8 pp, medium confidence |
| Flint | 74.2% | 68.4% | estimated ± 6.8 pp, medium confidence |
| MAI-Thinking-1 | 73.5% | 67.4% | estimated ± 6.8 pp, medium confidence |
| Jais-3-MoVA-36B-A4B | 72.8% | 66.5% | estimated ± 6.8 pp, medium confidence |
| Claude 4 Sonnet | 72.7% | 66.3% | estimated ± 6.8 pp, medium confidence |
| MAI-Code-1.1-Flash | 72.6% | 66.2% | estimated ± 6.8 pp, medium confidence |
| Claude 4 Opus | 72.5% | 66.0% | estimated ± 6.8 pp, medium confidence |
| Jais-3-32B | 72.4% | 65.9% | estimated ± 6.8 pp, medium confidence |
| Qwen3.5-27B | 72.4% | 65.9% | estimated ± 6.8 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B | 71.1% | 64.1% | estimated ± 6.8 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B NVFP4 | 71.1% | 64.1% | estimated ± 6.8 pp, medium confidence |
| Laguna XS 2.1 | 70.9% | 63.8% | estimated ± 6.8 pp, medium confidence |
| Grok Code Fast 1 | 70.8% | 63.7% | estimated ± 6.8 pp, medium confidence |
| Solar Pro 4 | 70.6% | 63.4% | estimated ± 6.8 pp, medium confidence |
| Solar Open 2 | 70.4% | 63.2% | estimated ± 6.8 pp, medium confidence |
| Jais-3-375B-A23B | 70.2% | 62.9% | estimated ± 6.8 pp, medium confidence |
| Jais-3-7B | 70.2% | 62.9% | estimated ± 6.8 pp, medium confidence |
| Qwen3-Coder 480B-A35B | 69.6% | 62.1% | estimated ± 6.8 pp, medium confidence |
| Qwen3.5-35B-A3B | 69.2% | 61.5% | estimated ± 6.8 pp, medium confidence |
| LongCat-Flash-Lite-Sparse | 68.2% | 60.1% | estimated ± 6.8 pp, medium confidence |
| DeepSeek V3.1 | 66.0% | 57.1% | estimated ± 6.8 pp, medium confidence |
| Kimi K2 | 65.8% | 56.8% | estimated ± 6.8 pp, medium confidence |
| Claude 3.7 Sonnet | 62.3% | 52.0% | estimated ± 6.8 pp, low confidence |
| Ternary Bonsai 2 27B | 60.8% | 50.0% | estimated ± 6.8 pp, low confidence |
| Nemotron 3 Super 100B | 60.5% | 49.5% | estimated ± 6.8 pp, low confidence |
| GLM-4.7-Flash | 59.2% | 47.8% | estimated ± 6.8 pp, low confidence |
| DeepSeek-R1-0528 | 57.6% | 45.6% | estimated ± 6.8 pp, low confidence |
| MiniMax M1 80k | 56.0% | 43.4% | estimated ± 6.8 pp, low confidence |
| GPT-4.1 | 54.6% | 41.4% | estimated ± 6.8 pp, low confidence |
| ZAYA1-74B-Preview | 53.2% | 39.5% | estimated ± 6.8 pp, low confidence |
| o3-mini | 49.3% | 34.1% | estimated ± 6.8 pp, low confidence |
| LLaDA2.2-flash | 49.3% | 34.1% | estimated ± 6.8 pp, low confidence |
| DeepSeek-R1 | 49.2% | 34.0% | estimated ± 6.8 pp, low confidence |
| Claude 3.5 Sonnet | 49.0% | 33.7% | estimated ± 6.8 pp, low confidence |
| Mellum2.1-12B-A2.5B-Thinking | 47.0% | 31.0% | estimated ± 6.8 pp, low confidence |
| Devstral Small | 46.8% | 30.7% | estimated ± 6.8 pp, low confidence |
| MiniCPM5-2B | 46.4% | 30.1% | estimated ± 6.8 pp, low confidence |
| Sarvam 105B | 45.0% | 28.2% | estimated ± 6.8 pp, low confidence |
| LFM3-1.7B | 43.4% | 26.0% | estimated ± 6.8 pp, low confidence |
| Claude 3.5 Haiku | 40.6% | 22.2% | estimated ± 6.8 pp, low confidence |
| Sarvam 30B | 34.0% | 13.1% | estimated ± 6.8 pp, low confidence |
Fitted on
| Model | SWE-bench Verified | Vals SWE-bench |
|---|---|---|
| Claude Fable 5 | 95.0% | 95.0% |
| Claude Haiku 4.5 | 73.3% | 66.6% |
| Claude Opus 4.8 | 88.6% | 88.6% |
| Claude Opus 5 | 96.0% | 97.0% |
| Claude Sonnet 4.6 | 79.6% | 77.4% |
| Claude Sonnet 5 | 85.2% | 79.6% |
| DeepSeek V4 Flash 0731 | 79.0% | 88.8% |
| DeepSeek V4 Pro 0813 | 80.6% | 96.4% |
| Gemini 2.5 Pro | 63.8% | 54.4% |
| GLM-4.7 | 73.8% | 69.4% |
| GPT-5.3 Codex | 85.0% | 78.0% |
| Grok 4.20 | 76.7% | 72.2% |
| Inkling | 77.6% | 77.6% |
| Inkling-Small | 80.2% | 82.2% |
| Kimi K2.6 | 80.2% | 76.2% |
| Laguna M.1 | 74.6% | 57.6% |
| Laguna XS.2 | 69.9% | 55.2% |
| MiniMax M3 | 80.5% | 75.0% |
| Mistral Medium 3.5 128B | 77.6% | 66.4% |
| Muse Spark | 77.4% | 74.4% |
| Nemotron 3 Ultra | 71.9% | 69.0% |
| Qwen3.6-27B | 77.2% | 70.0% |
| Qwen3.6 Plus | 78.8% | 73.4% |
| Qwen3.7 Max | 80.4% | 68.8% |