Calibration
Vals SWE-bench → OpenHarmony Bench
OpenHarmony Bench is estimated from Vals SWE-bench with a linear curve fitted on 11 models measured on both: y = 0.3134·x + 0.2921, R² = 0.52, cross-validated error 3.2 pp. It is used for 13 estimates.
| Estimated model | Vals SWE-bench | OpenHarmony Bench | Source |
|---|---|---|---|
| Claude Haiku 4.5 | 66.6% | 50.1% | estimated ± 3.2 pp, medium confidence |
| Claude Opus 4.7 | 82.0% | 54.9% | estimated ± 3.2 pp, high confidence |
| Claude Sonnet 4.6 | 77.4% | 53.5% | estimated ± 3.2 pp, high confidence |
| Composer 2.5 | 79.6% | 54.2% | estimated ± 3.2 pp, high confidence |
| Gemini 3.1 Flash-Lite | 62.8% | 48.9% | estimated ± 3.2 pp, medium confidence |
| Gemini 3 Flash | 75.0% | 52.7% | estimated ± 3.2 pp, high confidence |
| GPT-5.2-Codex | 72.4% | 51.9% | estimated ± 3.2 pp, high confidence |
| GPT-5.3 Codex | 78.0% | 53.7% | estimated ± 3.2 pp, high confidence |
| Grok 4.20 | 72.2% | 51.8% | estimated ± 3.2 pp, high confidence |
| Laguna M.1 | 57.6% | 47.3% | estimated ± 3.2 pp, medium confidence |
| Laguna XS.2 | 55.2% | 46.5% | estimated ± 3.2 pp, medium confidence |
| MiMo-V2.5 | 71.0% | 51.5% | estimated ± 3.2 pp, high confidence |
| Qwen3.5 Flash | 64.4% | 49.4% | estimated ± 3.2 pp, medium confidence |