Skip to content

Benchmark

pfgen-bench Japanese generation: Truthfulness

Higher scores are better.

Models measured
101
Scores recorded
156
Sources
1
IQ weight
1.36

Score distribution

0.7
Mean
0.734
Median
0.215
Std dev
0.392
10th percentile
0.959
90th percentile
0.001
Min
0.0010.988
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.36 for chat models, measured over 101 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

Showing 151–156 of 156 results.

Data credits: pfgen-bench, Preferred Networks. Apache-2.0 except LLM-generated outputs; aggregate scores only. (licence)