Calibration
SWE-bench Verified → AA Coding Index
AA Coding Index is estimated from SWE-bench Verified with a offset logistic curve fitted on 35 models measured on both: y = 0.2336 + (0.8143 − 0.2336) / (1 + exp(−14.88·(x − 0.7611))), R² = 0.91, cross-validated error 5.3 pp. It is used for 61 estimates.
| Estimated model | SWE-bench Verified | AA Coding Index | Source |
|---|---|---|---|
| Claude Mythos 5 | 95.5% | 78.4% | estimated ± 5.3 pp, medium confidence |
| GPT-5.3 Codex | 85.0% | 69.2% | estimated ± 5.3 pp, medium confidence |
| GPT-5.2 | 80.0% | 60.6% | estimated ± 5.3 pp, medium confidence |
| BTL-4 | 78.4% | 57.3% | estimated ± 5.3 pp, medium confidence |
| MiMo-V2-Pro | 78.0% | 56.4% | estimated ± 5.3 pp, medium confidence |
| Claude Sonnet 4.5 | 77.2% | 54.7% | estimated ± 5.3 pp, medium confidence |
| Grok 4.20 | 76.7% | 53.7% | estimated ± 5.3 pp, medium confidence |
| MiMo-V2-Omni | 74.8% | 49.6% | estimated ± 5.3 pp, medium confidence |
| Laguna M.1 | 74.6% | 49.1% | estimated ± 5.3 pp, medium confidence |
| Claude 4.1 Opus | 74.5% | 48.9% | estimated ± 5.3 pp, medium confidence |
| Step 3.5 Flash | 74.4% | 48.7% | estimated ± 5.3 pp, medium confidence |
| Flint | 74.2% | 48.3% | estimated ± 5.3 pp, medium confidence |
| MAI-Thinking-1 | 73.5% | 46.8% | estimated ± 5.3 pp, medium confidence |
| Claude Haiku 4.5 | 73.3% | 46.4% | estimated ± 5.3 pp, medium confidence |
| Jais-3-MoVA-36B-A4B | 72.8% | 45.4% | estimated ± 5.3 pp, medium confidence |
| Claude 4 Sonnet | 72.7% | 45.2% | estimated ± 5.3 pp, medium confidence |
| MAI-Code-1.1-Flash | 72.6% | 45.0% | estimated ± 5.3 pp, medium confidence |
| Claude 4 Opus | 72.5% | 44.8% | estimated ± 5.3 pp, medium confidence |
| Jais-3-32B | 72.4% | 44.6% | estimated ± 5.3 pp, medium confidence |
| Qwen3.5-27B | 72.4% | 44.6% | estimated ± 5.3 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B | 71.1% | 42.0% | estimated ± 5.3 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B NVFP4 | 71.1% | 42.0% | estimated ± 5.3 pp, medium confidence |
| Laguna XS 2.1 | 70.9% | 41.7% | estimated ± 5.3 pp, medium confidence |
| Grok Code Fast 1 | 70.8% | 41.5% | estimated ± 5.3 pp, medium confidence |
| Solar Pro 4 | 70.6% | 41.1% | estimated ± 5.3 pp, medium confidence |
| Solar Open 2 | 70.4% | 40.7% | estimated ± 5.3 pp, medium confidence |
| Jais-3-375B-A23B | 70.2% | 40.4% | estimated ± 5.3 pp, medium confidence |
| Jais-3-7B | 70.2% | 40.4% | estimated ± 5.3 pp, medium confidence |
| Laguna XS.2 | 69.9% | 39.9% | estimated ± 5.3 pp, medium confidence |
| Qwen3-Coder 480B-A35B | 69.6% | 39.3% | estimated ± 5.3 pp, medium confidence |
| Qwen3.5-35B-A3B | 69.2% | 38.6% | estimated ± 5.3 pp, medium confidence |
| K-EXAONE 2.0 | 68.2% | 37.0% | estimated ± 5.3 pp, medium confidence |
| LongCat-Flash-Lite-Sparse | 68.2% | 37.0% | estimated ± 5.3 pp, medium confidence |
| DeepSeek V3.1 | 66.0% | 33.9% | estimated ± 5.3 pp, medium confidence |
| Kimi K2 | 65.8% | 33.7% | estimated ± 5.3 pp, medium confidence |
| Spark-X2.6-4B | 62.8% | 30.4% | estimated ± 5.3 pp, medium confidence |
| Claude 3.7 Sonnet | 62.3% | 29.9% | estimated ± 5.3 pp, medium confidence |
| Ternary Bonsai 2 27B | 60.8% | 28.8% | estimated ± 5.3 pp, medium confidence |
| Nemotron 3 Super 100B | 60.5% | 28.5% | estimated ± 5.3 pp, medium confidence |
| GLM-4.7-Flash | 59.2% | 27.7% | estimated ± 5.3 pp, medium confidence |
| DeepSeek-R1-0528 | 57.6% | 26.8% | estimated ± 5.3 pp, medium confidence |
| Granite 4.2 30B | 57.0% | 26.6% | estimated ± 5.3 pp, medium confidence |
| MiniMax M1 80k | 56.0% | 26.1% | estimated ± 5.3 pp, medium confidence |
| GPT-4.1 | 54.6% | 25.6% | estimated ± 5.3 pp, medium confidence |
| ZAYA1-74B-Preview | 53.2% | 25.2% | estimated ± 5.3 pp, medium confidence |
| LFM2.5-24B-A2B | 51.0% | 24.7% | estimated ± 5.3 pp, medium confidence |
| o3-mini | 49.3% | 24.4% | estimated ± 5.3 pp, medium confidence |
| LLaDA2.2-flash | 49.3% | 24.4% | estimated ± 5.3 pp, medium confidence |
| DeepSeek-R1 | 49.2% | 24.4% | estimated ± 5.3 pp, medium confidence |
| Claude 3.5 Sonnet | 49.0% | 24.4% | estimated ± 5.3 pp, medium confidence |
| Mellum2.1-12B-A2.5B-Thinking | 47.0% | 24.1% | estimated ± 5.3 pp, medium confidence |
| Devstral Small | 46.8% | 24.1% | estimated ± 5.3 pp, medium confidence |
| MiniCPM5-2B | 46.4% | 24.0% | estimated ± 5.3 pp, medium confidence |
| Sarvam 105B | 45.0% | 23.9% | estimated ± 5.3 pp, medium confidence |
| LFM3-1.7B | 43.4% | 23.8% | estimated ± 5.3 pp, medium confidence |
| Claude 3.5 Haiku | 40.6% | 23.6% | estimated ± 5.3 pp, medium confidence |
| Sarvam 30B | 34.0% | 23.5% | estimated ± 5.3 pp, medium confidence |
| Gemini 2.5 Flash-Lite | 31.6% | 23.4% | estimated ± 5.3 pp, medium confidence |
| Gemini 2.0 Flash | 21.4% | 23.4% | estimated ± 5.3 pp, low confidence |
| Gemini 2.0 Flash-Lite | 12.5% | 23.4% | estimated ± 5.3 pp, low confidence |
| Gemini 1.5 Flash | 9.6% | 23.4% | estimated ± 5.3 pp, low confidence |
Fitted on
| Model | SWE-bench Verified | AA Coding Index |
|---|---|---|
| Apodex 1.1 | 77.7% | 60.8% |
| Claude Fable 5 | 95.0% | 76.5% |
| Claude Opus 4.7 (Adaptive) | 87.6% | 73.6% |
| Claude Opus 4.8 | 88.6% | 74.3% |
| Claude Opus 5 | 96.0% | 78.0% |
| Claude Sonnet 5 | 85.2% | 71.6% |
| DeepSeek V3 | 42.0% | 23.0% |
| DeepSeek V4 Flash 0731 | 79.0% | 69.1% |
| DeepSeek V4 Pro 0813 | 80.6% | 68.8% |
| Gemini 2.5 Pro | 63.8% | 33.3% |
| GLM-4.7 | 73.8% | 45.3% |
| GPT-4.1 mini | 23.6% | 20.2% |
| GPT-OSS 120B | 62.4% | 30.4% |
| GPT-OSS 20B | 60.7% | 20.7% |
| Granite 4.2 8B | 47.7% | 22.4% |
| Hy3 Preview | 74.4% | 58.8% |
| Inkling | 77.6% | 52.1% |
| Inkling-Small | 80.2% | 52.9% |
| K-Exaone | 49.4% | 32.1% |
| Kimi K2.6 | 80.2% | 61.8% |
| Kimi K2.5 | 76.8% | 46.8% |
| Kimi K2.5 (Reasoning) | 76.8% | 46.8% |
| MiMo-V2-Flash | 73.4% | 49.8% |
| MiniMax M3 | 80.5% | 58.6% |
| Mistral Medium 3.5 128B | 77.6% | 46.9% |
| Muse Glimmer 30B | 76.0% | 49.0% |
| Muse Spark | 77.4% | 58.6% |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 52.8% | 26.8% |
| Nemotron 3 Ultra | 71.9% | 49.3% |
| Qwen3.5-122B-A10B | 72.0% | 45.7% |
| Qwen3.6-27B | 77.2% | 53.7% |
| Qwen3.6-35B-A3B | 73.4% | 41.9% |
| Qwen3.6 Plus | 78.8% | 54.5% |
| Qwen3.7 Max | 80.4% | 66.0% |
| Qwen3.7 Plus | 77.7% | 55.9% |