Calibration
SWE-bench Pro → OpenHarmony Bench
OpenHarmony Bench is estimated from SWE-bench Pro with a offset logistic curve fitted on 7 models measured on both: y = 0.5328 + (0.6090 − 0.5328) / (1 + exp(−200.00·(x − 0.6181))), R² = 0.50, cross-validated error 4.9 pp. It is used for 14 estimates.
| Estimated model | SWE-bench Pro | OpenHarmony Bench | Source |
|---|---|---|---|
| Claude Mythos 5 | 80.3% | 60.9% | estimated ± 4.9 pp, low confidence |
| Sakana Fugu-Ultra | 73.7% | 60.9% | estimated ± 4.9 pp, low confidence |
| Beam 1.0 501B-A23B FP8 | 65.9% | 60.9% | estimated ± 4.9 pp, low confidence |
| Beam 1.0 501B-A23B NVFP4 | 65.6% | 60.9% | estimated ± 4.9 pp, low confidence |
| Sage-A1.5 | 61.2% | 55.0% | estimated ± 4.9 pp, low confidence |
| Atria Dawn Preview | 59.6% | 53.4% | estimated ± 4.9 pp, low confidence |
| Laguna S 2.1 | 59.4% | 53.3% | estimated ± 4.9 pp, low confidence |
| Sakana Fugu | 59.0% | 53.3% | estimated ± 4.9 pp, low confidence |
| GPT-5.2 | 55.6% | 53.3% | estimated ± 4.9 pp, low confidence |
| Spark-X2.6-4B | 54.3% | 53.3% | estimated ± 4.9 pp, low confidence |
| Claude Opus 4.6 | 53.4% | 53.3% | estimated ± 4.9 pp, low confidence |
| MAI-Thinking-1 | 52.8% | 53.3% | estimated ± 4.9 pp, low confidence |
| Laguna XS 2.1 | 47.6% | 53.3% | estimated ± 4.9 pp, low confidence |
| LongCat-Flash-Lite-Sparse | 40.6% | 53.3% | estimated ± 4.9 pp, low confidence |
Fitted on
| Model | SWE-bench Pro | OpenHarmony Bench |
|---|---|---|
| DeepSeek V4 Flash 0731 | 52.6% | 53.8% |
| DeepSeek V4 Pro 0813 | 55.4% | 59.0% |
| GLM-5.1 | 58.4% | 52.3% |
| GLM-5.2 | 62.1% | 58.4% |
| MiniMax M3 | 59.0% | 48.4% |
| Qwen3.7 Max | 60.6% | 53.4% |
| Qwen3.8 Max | 67.7% | 60.8% |