Skip to content

Benchmark

AHELM: Toxicity detection - Mean win rate

Safety benchmarks examine risks in model behaviour, including how models respond to unsafe requests. Scores are percentages. Higher is better.

Models measured
12
Scores recorded
12
Sources
1
IQ weight
0.28

Score distribution

52.1%
Mean
62.5%
Median
31.3%
Std dev
12.5%
10th percentile
93.8%
90th percentile
6.3%
Min
6.3%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.28 for chat models, measured over 9 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard