benchgap
Calibration

Decision Index Language → When2Call (Decision Index)

When2Call (Decision Index) is estimated from Decision Index Language with a Michaelis–Menten curve fitted on 11 models measured on both: y = 1.7269·x / (0.60203 + x), R² = 0.52, cross-validated error 9.4 pp. It is used for 1 estimate.

Estimated modelDecision Index LanguageWhen2Call (Decision Index)Source
Tev1-0.8B-experimental14.4%33.3%estimated ± 9.4 pp, low confidence

Fitted on

ModelDecision Index LanguageWhen2Call (Decision Index)
Clef47.8%71.0%
Clef-Flash40.5%64.4%
Clef-Omni40.7%62.1%
Drex Max62.1%94.0%
Drex v1.540.8%84.8%
Drex v2.052.7%86.2%
elephant-162.1%94.8%
GLiDE v1.266.8%94.4%
Jev 1.1458.0%80.8%
Perplexity Decider Flash v1.2 4B62.2%74.4%
Perplexity Decider v1.2 27B65.5%87.7%