Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Sexual Content

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

Models measured
70
Scores recorded
75
Sources
1
IQ weight
0.48

Score distribution

55.8%
Mean
50.7%
Median
16.3%
Std dev
38.5%
10th percentile
80.2%
90th percentile
24.7%
Min
24.7%95.1%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.48 for chat models, measured over 70 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs95.1%
2Claude Haiku 4.5· Stanford HELM specialized suites data docs87.5%
3Palmyra X5· Stanford HELM specialized suites data docs85.4%
4Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs83%
5Claude 3 Sonnet· Stanford HELM specialized suites data docs82.6%
6Claude Sonnet 4· Stanford HELM specialized suites data docs81.9%
7Claude 3 Haiku· Stanford HELM specialized suites data docs80.6%
8Llama3 8B Chat· Stanford HELM specialized suites data docs80.2%
9Claude Opus 4· Stanford HELM specialized suites data docs79.2%
10GPT OSS 20b· Stanford HELM specialized suites data docs78.8%
11Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs77.8%
12Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs75.7%
13Claude Sonnet 3.7· Stanford HELM specialized suites data docs75.3%
14GPT OSS 120B· Stanford HELM specialized suites data docs75.3%
15Granite 4.0 H Small· Stanford HELM specialized suites data docs74%
16GPT 5.1· Stanford HELM specialized suites data docs72.6%
17Claude Sonnet 4.5· Stanford HELM specialized suites data docs71.9%
18Llama 4 Maverick· Stanford HELM specialized suites data docs71.9%
19Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs70.5%
20Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs69.4%
21gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs69.1%
22Claude 3 Opus· Stanford HELM specialized suites data docs68.8%
23Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs68.1%
24Gemini 2.0 Flash· Stanford HELM specialized suites data docs67%
25GPT-4 Turbo· Stanford HELM specialized suites data docs65.3%
26GPT-4.5 Preview· Stanford HELM specialized suites data docs61.5%
27Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs61.1%
28GPT 5· Stanford HELM specialized suites data docs59.4%
29GPT-5 mini· Stanford HELM specialized suites data docs55.2%
30GPT-5 nano· Stanford HELM specialized suites data docs55.2%
31Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs55.2%
32Llama 4 Scout· Stanford HELM specialized suites data docs54.2%
33Qwen2 72B Instruct· Stanford HELM specialized suites data docs53.8%
34GPT 4· Stanford HELM specialized suites data docs52.4%
35GPT 4osetting not named by the source· Stanford HELM specialized suites data docs52.1%
36GPT-4.1 nano· Stanford HELM specialized suites data docs51.7%
37gemini-1.5-flash-002· Stanford HELM specialized suites data docs51%
38GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs50.7%
39OpenAI o3· Stanford HELM specialized suites data docs50.7%
40GPT 4osetting not named by the source· Stanford HELM specialized suites data docs50%
41Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs49.3%
42GPT 4.1· Stanford HELM specialized suites data docs49.3%
43gemini-1.5-pro-002· Stanford HELM specialized suites data docs49%
44GPT-4.1 mini· Stanford HELM specialized suites data docs47.9%
45Mistral Large2407· Stanford HELM specialized suites data docs47.2%
46o4 Mini· Stanford HELM specialized suites data docs46.9%
47GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs46.5%
48GPT-4o mini· Stanford HELM specialized suites data docs46.5%
49Kimi K2 Instruct· Stanford HELM specialized suites data docs46.5%
50Yi 34B Chat· Stanford HELM specialized suites data docs46.2%

Showing 1–50 of 75 results.

Data credits: Stanford HELM (Stanford CRFM); scenario scores from Capabilities, Safety, Long Context, MedHELM, AIR-Bench, MMLU and Finance