Calibration
AA Coding Index → Vals SWE-bench
Vals SWE-bench is estimated from AA Coding Index with a linear curve fitted on 44 models measured on both: y = 0.7381·x + 0.3338, R² = 0.70, cross-validated error 5.4 pp. It is used for 41 estimates.
| Estimated model | AA Coding Index | Vals SWE-bench | Source |
|---|---|---|---|
| Apodex 1.1 | 60.8% | 78.2% | estimated ± 5.4 pp, medium confidence |
| Apodex 1.1 Mini | 60.8% | 78.2% | estimated ± 5.4 pp, medium confidence |
| Celeris-1 | 14.4% | 44.0% | estimated ± 5.4 pp, low confidence |
| Claude 3 Opus | 19.5% | 47.8% | estimated ± 5.4 pp, low confidence |
| Claude Opus 4.7 (Adaptive) | 73.6% | 87.7% | estimated ± 5.4 pp, medium confidence |
| Command A+ | 27.9% | 53.9% | estimated ± 5.4 pp, low confidence |
| DeepSeek V3 | 23.0% | 50.4% | estimated ± 5.4 pp, low confidence |
| Gemini 1.5 Pro | 23.6% | 50.8% | estimated ± 5.4 pp, low confidence |
| Gemma 3 27B | 10.1% | 40.8% | estimated ± 5.4 pp, low confidence |
| Gemma 4 12B | 31.0% | 56.2% | estimated ± 5.4 pp, low confidence |
| Gemma 4 26B A4B | 39.3% | 62.4% | estimated ± 5.4 pp, medium confidence |
| Gemma 4 E2B | 7.2% | 38.7% | estimated ± 5.4 pp, low confidence |
| Gemma 4 E4B | 9.4% | 40.3% | estimated ± 5.4 pp, low confidence |
| GPT-4.1 mini | 20.2% | 48.3% | estimated ± 5.4 pp, low confidence |
| GPT-4.1 nano | 11.1% | 41.6% | estimated ± 5.4 pp, low confidence |
| GPT-4 Turbo | 21.5% | 49.2% | estimated ± 5.4 pp, low confidence |
| GPT-4o mini | 11.4% | 41.8% | estimated ± 5.4 pp, low confidence |
| GPT-5.1 | 49.4% | 69.8% | estimated ± 5.4 pp, medium confidence |
| GPT-5 (high) | 37.8% | 61.3% | estimated ± 5.4 pp, medium confidence |
| Hy3 | 58.8% | 76.8% | estimated ± 5.4 pp, medium confidence |
| K-Exaone | 32.1% | 57.1% | estimated ± 5.4 pp, low confidence |
| Kimi K2.5 (Reasoning) | 46.8% | 67.9% | estimated ± 5.4 pp, medium confidence |
| LFM2.5-2.6B | 7.7% | 39.1% | estimated ± 5.4 pp, low confidence |
| Llama 4 Maverick | 16.3% | 45.4% | estimated ± 5.4 pp, low confidence |
| Llama 4 Scout | 8.2% | 39.4% | estimated ± 5.4 pp, low confidence |
| MiMo-V2-Flash | 49.8% | 70.2% | estimated ± 5.4 pp, medium confidence |
| Mistral Large 3 | 20.1% | 48.2% | estimated ± 5.4 pp, low confidence |
| Mistral Small 4 | 26.6% | 53.0% | estimated ± 5.4 pp, low confidence |
| Mistral Small 4 (Reasoning) | 26.6% | 53.0% | estimated ± 5.4 pp, low confidence |
| Muse Spark 1.3 | 75.8% | 89.3% | estimated ± 5.4 pp, medium confidence |
| Nemotron 3 Nano 30B | 14.4% | 44.0% | estimated ± 5.4 pp, low confidence |
| Nemotron 3 Super 100B | 37.7% | 61.2% | estimated ± 5.4 pp, medium confidence |
| o1 | 39.7% | 62.7% | estimated ± 5.4 pp, medium confidence |
| o1-preview | 34.1% | 58.5% | estimated ± 5.4 pp, medium confidence |
| Quasar 438B | 61.2% | 78.5% | estimated ± 5.4 pp, medium confidence |
| Qwen3.5-122B-A10B | 45.7% | 67.1% | estimated ± 5.4 pp, medium confidence |
| Qwen3.8 Max Preview | 71.8% | 86.4% | estimated ± 5.4 pp, medium confidence |
| Step 3.7 Flash | 39.6% | 62.6% | estimated ± 5.4 pp, medium confidence |
| Trinity-Large-Preview | 25.8% | 52.4% | estimated ± 5.4 pp, low confidence |
| Trinity-Large-Thinking | 25.8% | 52.4% | estimated ± 5.4 pp, low confidence |
| Ultravox v0.6 Llama 3.3 70B | 11.9% | 42.2% | estimated ± 5.4 pp, low confidence |