Calibration
AA Coding Index → SWE-bench Pro
SWE-bench Pro is estimated from AA Coding Index with a Michaelis–Menten curve fitted on 38 models measured on both: y = 1.3899·x / (0.83100 + x), R² = 0.68, cross-validated error 6.0 pp. It is used for 36 estimates.
| Estimated model | AA Coding Index | SWE-bench Pro | Source |
|---|---|---|---|
| Apodex 1.1 Mini | 60.8% | 58.7% | estimated ± 6.0 pp, medium confidence |
| Celeris-1 | 14.4% | 20.5% | estimated ± 6.0 pp, low confidence |
| Claude 3 Opus | 19.5% | 26.4% | estimated ± 6.0 pp, low confidence |
| Command A+ | 27.9% | 34.9% | estimated ± 6.0 pp, medium confidence |
| Gemini 1.5 Pro | 23.6% | 30.8% | estimated ± 6.0 pp, medium confidence |
| Gemma 3 27B | 10.1% | 15.0% | estimated ± 6.0 pp, low confidence |
| Gemma 4 26B A4B | 39.3% | 44.6% | estimated ± 6.0 pp, medium confidence |
| Gemma 4 31B | 43.4% | 47.7% | estimated ± 6.0 pp, medium confidence |
| Gemma 4 E2B | 7.2% | 11.1% | estimated ± 6.0 pp, low confidence |
| Gemma 4 E4B | 9.4% | 14.1% | estimated ± 6.0 pp, low confidence |
| GPT-4.1 nano | 11.1% | 16.4% | estimated ± 6.0 pp, low confidence |
| GPT-4 Turbo | 21.5% | 28.6% | estimated ± 6.0 pp, low confidence |
| GPT-4o mini | 11.4% | 16.7% | estimated ± 6.0 pp, low confidence |
| GPT-5.1 | 49.4% | 51.8% | estimated ± 6.0 pp, medium confidence |
| GPT-5 (high) | 37.8% | 43.4% | estimated ± 6.0 pp, medium confidence |
| GPT-OSS 120B | 30.4% | 37.3% | estimated ± 6.0 pp, medium confidence |
| GPT-OSS 20B | 20.7% | 27.7% | estimated ± 6.0 pp, low confidence |
| Hy3 | 58.8% | 57.6% | estimated ± 6.0 pp, medium confidence |
| K-Exaone | 32.1% | 38.7% | estimated ± 6.0 pp, medium confidence |
| Ling 2.6 Flash | 25.3% | 32.4% | estimated ± 6.0 pp, medium confidence |
| Ling 3.0 Flash FP8 | 50.6% | 52.6% | estimated ± 6.0 pp, medium confidence |
| Llama 4 Maverick | 16.3% | 22.8% | estimated ± 6.0 pp, low confidence |
| Llama 4 Scout | 8.2% | 12.4% | estimated ± 6.0 pp, low confidence |
| Mistral Large 3 | 20.1% | 27.0% | estimated ± 6.0 pp, low confidence |
| Mistral Small 4 | 26.6% | 33.7% | estimated ± 6.0 pp, medium confidence |
| Mistral Small 4 (Reasoning) | 26.6% | 33.7% | estimated ± 6.0 pp, medium confidence |
| Nemotron 3 Nano 30B | 14.4% | 20.5% | estimated ± 6.0 pp, low confidence |
| Nemotron 3 Nano Omni 30B A3B | 13.8% | 19.7% | estimated ± 6.0 pp, low confidence |
| Nemotron 3 Super 100B | 37.7% | 43.4% | estimated ± 6.0 pp, medium confidence |
| o1 | 39.7% | 44.9% | estimated ± 6.0 pp, medium confidence |
| o1-preview | 34.1% | 40.4% | estimated ± 6.0 pp, medium confidence |
| Quasar 438B | 61.2% | 58.9% | estimated ± 6.0 pp, medium confidence |
| Qwen3.8 Max Preview | 71.8% | 64.4% | estimated ± 6.0 pp, medium confidence |
| Trinity-Large-Preview | 25.8% | 32.9% | estimated ± 6.0 pp, medium confidence |
| Trinity-Large-Thinking | 25.8% | 32.9% | estimated ± 6.0 pp, medium confidence |
| Ultravox v0.6 Llama 3.3 70B | 11.9% | 17.4% | estimated ± 6.0 pp, low confidence |