benchgap
Calibration

LiveCodeBench v6 → NL2Repo

NL2Repo is estimated from LiveCodeBench v6 with a Michaelis–Menten curve fitted on 5 models measured on both: y = 2.0000·x / (2.97945 + x), R² = 0.37, cross-validated error 3.1 pp. It is used for 26 estimates.

Estimated modelLiveCodeBench v6NL2RepoSource
A.X K284.0%44.0%estimated ± 3.1 pp, low confidence
Agents-A1-4B59.6%33.3%estimated ± 3.1 pp, low confidence
BTL-388.1%45.6%estimated ± 3.1 pp, low confidence
BTL-466.1%36.3%estimated ± 3.1 pp, low confidence
Gemma 4 12B72.0%38.9%estimated ± 3.1 pp, low confidence
Granite 4.2 30B75.8%40.5%estimated ± 3.1 pp, low confidence
Granite 4.2 3B69.7%37.9%estimated ± 3.1 pp, low confidence
Granite 4.2 8B73.2%39.5%estimated ± 3.1 pp, low confidence
Kimi K2.689.6%46.2%estimated ± 3.1 pp, low confidence
Kimi K2.585.0%44.4%estimated ± 3.1 pp, low confidence
LFM2.5-2.6B59.4%33.2%estimated ± 3.1 pp, low confidence
LLaDA2.2-mini28.1%17.3%estimated ± 3.1 pp, low confidence
MAI-Thinking-187.7%45.5%estimated ± 3.1 pp, low confidence
Mellum2-12B-A2.5B-Instruct37.2%22.2%estimated ± 3.1 pp, low confidence
Mellum2-12B-A2.5B-Thinking69.9%38.0%estimated ± 3.1 pp, low confidence
MiniCPM5-1B33.5%20.2%estimated ± 3.1 pp, low confidence
MiniCPM5-2B69.1%37.7%estimated ± 3.1 pp, low confidence
Nemotron 3 Ultra89.0%46.0%estimated ± 3.1 pp, low confidence
Qwen3.5 397B83.6%43.8%estimated ± 3.1 pp, low confidence
Qwen3.6 Plus87.1%45.2%estimated ± 3.1 pp, low confidence
Sakana Fugu92.9%47.5%estimated ± 3.1 pp, low confidence
Sakana Fugu-Ultra93.2%47.7%estimated ± 3.1 pp, low confidence
Solar Open 292.4%47.3%estimated ± 3.1 pp, low confidence
Ternary Bonsai 2 27B90.1%46.4%estimated ± 3.1 pp, low confidence
ZAYA1-74B-Preview65.7%36.1%estimated ± 3.1 pp, low confidence
ZAYA1-8B65.8%36.2%estimated ± 3.1 pp, low confidence