Skip to content

Benchmark

PashtoBench safety refusal rate on sensitive prompts (English)

Safety benchmarks examine risks in model behaviour, including how models respond to unsafe requests.

Models measured
4
Scores recorded
4
Sources
1
IQ weight
No weight published

Score distribution

0.847
Mean
0.842
Median
0.067
Std dev
0.773
10th percentile
0.955
90th percentile
0.773
Min
0.7730.955
The band covers the middle 80% of scores. The line marks the median.

Results

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

ModelScore
Claude Sonnet 5column: refusal sens en· PashtoBench v1.0 data docs0.955
GPT 4ocolumn: refusal sens en· PashtoBench v1.0 data docs0.818
Llama 3.3 70B Instructcolumn: refusal sens en· PashtoBench v1.0 data docs0.773
Qwen2.5 72B Instructcolumn: refusal sens en· PashtoBench v1.0 data docs0.842

Data credits: PashtoBench v1.0, Jawad Noori. Public results/scores numbers; code MIT and data/ CC BY 4.0 grants do not name results/, so no results licence is asserted.