Skip to content

Benchmark

ChemBench Toxicity and Safety

Science benchmarks measure how well models answer science questions. Higher scores are better.

Models measured
42
Scores recorded
51
Sources
1
IQ weight
1.49

Score distribution

0.36
Mean
0.373
Median
0.095
Std dev
0.267
10th percentile
0.474
90th percentile
0.0148
Min
0.01480.516
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.49 for chat models, measured over 42 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51Llama-2-7b-chat-hfLlama-2-7b-chat-hf -base-1741273893· ChemBench verified leaderboard data docs0.0148

Showing 51–51 of 51 results.

Data credits: ChemBench leaderboard, Jablonka group / lamalab-org. Public verified numeric result summaries; results licence not separately stated.