Benchmark
AHELM: Air-Bench Chat (knowledge subsets) - GPT-4o Judge Critique
Safety benchmarks examine risks in model behaviour, including how models respond to unsafe requests. Scores are percentages.
- Models measured
- 12
- Scores recorded
- 36
- Sources
- 1
- IQ weight
- 2.62
Score distribution
- 2.9%
- Mean
- 2.9%
- Median
- 0.316%
- Std dev
- 2.4%
- 10th percentile
- 3.3%
- 90th percentile
- 2.2%
- Min
This benchmark has a weight of 2.62 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.