Skip to content

Benchmark

First, Do NOHARM (v2)

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

Models measured
13
Scores recorded
13
Sources
1
IQ weight
1.16

Score distribution

66.2%
Mean
65%
Median
7%
Std dev
57.9%
10th percentile
74.6%
90th percentile
55.8%
Min
55.8%79.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.16 for chat models, measured over 13 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Muse Spark 1.1clinical-result-288· Clinical Benchmarks data docs79.7%
2Claude Opus 5clinical-result-124· Clinical Benchmarks data docs74.6%
3Kimi K3clinical-result-125· Clinical Benchmarks data docs74%
4GPT-5.6 Solclinical-result-126· Clinical Benchmarks data docs70.1%
5GPT 5.5clinical-result-127· Clinical Benchmarks data docs70%
6GPT 5clinical-result-294· Clinical Benchmarks data docs68.6%
7Claude Fable 5clinical-result-289· Clinical Benchmarks data docs65%
8Gemini 3.1 Pro Previewclinical-result-128· Clinical Benchmarks data docs62.6%
9Gemini 2.5 Proclinical-result-290· Clinical Benchmarks data docs61.9%
10Qwen3.5 397B A17Bclinical-result-291· Clinical Benchmarks data docs61.1%
11Kimi K2.6clinical-result-292· Clinical Benchmarks data docs59.1%
12GLM 5.1clinical-result-1821· Clinical Benchmarks data docs57.9%
13DeepSeek R1clinical-result-293· Clinical Benchmarks data docs55.8%

Data credits: Clinical Benchmarks (clinicalbenchmarks.ai), CC BY 4.0; fractions converted to percentages; source documents retain their own terms (licence)