Calibration
NL2Repo → LiveCodeBench v6
LiveCodeBench v6 is estimated from NL2Repo with a Michaelis–Menten curve fitted on 5 models measured on both: y = 1.3254·x / (0.21727 + x), R² = 0.48, cross-validated error 3.8 pp. It is used for 2 estimates.
| Estimated model | NL2Repo | LiveCodeBench v6 | Source |
|---|---|---|---|
| Seed 2.1 Pro | 47.0% | 90.6% | estimated ± 3.8 pp, low confidence |
| Seed 2.1 Turbo | 43.7% | 88.5% | estimated ± 3.8 pp, low confidence |