/

← All benchmarks

CRUXEval

coding · code-reasoning↗ Quelle

48Evidence

not yet measured

What it measures

Ob ein Modell Ein- und Ausgabe von Code vorhersagen kann, ohne ihn auszufuehren.

What it does not measure

Codeerzeugung.

How the score is composed

Currency50

Discrimination50

Headroom50

Contamination safety33

Gaming resistance55

Methodological rigor70

Reproducibility60

Details

Benchmaxxing risk: medium
Difficulty: medium
Public test set: yes
Human-verified: no
LLM-judged: no
Sample size: 800
Skala: 0–100 % correct (höher besser)

Best measured models

No results ingested yet.