benchgap
Calibration

NL2Repo → OpenHarmony Bench

OpenHarmony Bench is estimated from NL2Repo with a Michaelis–Menten + offset curve fitted on 10 models measured on both: y = 0.0000 + 0.9654·x / (0.37363 + x), R² = 0.70, cross-validated error 2.7 pp. It is used for 13 estimates.

Estimated modelNL2RepoOpenHarmony BenchSource
Claude Opus 4.543.2%51.8%estimated ± 2.7 pp, high confidence
dots3-note Preview49.8%55.2%estimated ± 2.7 pp, high confidence
Hy4 preview58.9%59.1%estimated ± 2.7 pp, high confidence
Ornith-1.0-35B34.6%46.4%estimated ± 2.7 pp, medium confidence
Ornith-1.0-397B48.2%54.4%estimated ± 2.7 pp, high confidence
Ornith-1.0-9B27.2%40.7%estimated ± 2.7 pp, medium confidence
Ornith-1.5-35B-A3B46.2%53.4%estimated ± 2.7 pp, high confidence
Ornith-1.5-397B59.5%59.3%estimated ± 2.7 pp, high confidence
Ornith-1.5-9B32.4%44.8%estimated ± 2.7 pp, medium confidence
Qwen 3.6 Max (preview)42.9%51.6%estimated ± 2.7 pp, high confidence
Qwen3.8-Omni-Flash48.9%54.7%estimated ± 2.7 pp, high confidence
Seed 2.1 Pro47.0%53.8%estimated ± 2.7 pp, high confidence
Seed 2.1 Turbo43.7%52.0%estimated ± 2.7 pp, high confidence