Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #17: Adult 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
0.00

Score distribution

65.6%
Mean
65.2%
Median
14.2%
Std dev
46.4%
10th percentile
83.3%
90th percentile
27.5%
Min
27.5%96.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.00 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 docs96.4%
2Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs88.4%
3Claude Haiku 4.5· Stanford HELM specialized suites data docs87.7%
4Claude Sonnet 4· Stanford HELM specialized suites data docs86.2%
5Claude Sonnet 3.7· Stanford HELM specialized suites data docs85.5%
6Claude Opus 4· Stanford HELM specialized suites data docs84.1%
7Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs83.3%
8Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs83.3%
9GPT-4 Turbo· Stanford HELM specialized suites data docs82.6%
10Gemini 2.0 Flash· Stanford HELM specialized suites data docs81.9%
11GPT OSS 20b· Stanford HELM specialized suites data docs81.2%
12Palmyra X5· Stanford HELM specialized suites data docs81.2%
13GPT OSS 120B· Stanford HELM specialized suites data docs80.4%
14Granite 4.0 H Small· Stanford HELM specialized suites data docs80.4%
15Llama3 8B Chat· Stanford HELM specialized suites data docs80.4%
16Claude 3 Sonnet· Stanford HELM specialized suites data docs78.3%
17Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs78.3%
18Claude Sonnet 4.5· Stanford HELM specialized suites data docs76.1%
19gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs76.1%
20Llama 4 Maverick· Stanford HELM specialized suites data docs75.4%
21Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs75.4%
22Claude 3 Haiku· Stanford HELM specialized suites data docs74.6%
23Llama 4 Scout· Stanford HELM specialized suites data docs74.6%
24Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs74.6%
25GPT 4osetting not named by the source· Stanford HELM specialized suites data docs73.9%
26GPT-5 nano· Stanford HELM specialized suites data docs73.9%
27GPT-4.5 Preview· Stanford HELM specialized suites data docs72.5%
28Mistral Large2407· Stanford HELM specialized suites data docs71.7%
29GPT 5.1· Stanford HELM specialized suites data docs71%
30GPT 4· Stanford HELM specialized suites data docs70.3%
31Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs70.3%
32Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs70.3%
33GPT 4osetting not named by the source· Stanford HELM specialized suites data docs68.8%
34GPT 5· Stanford HELM specialized suites data docs68.8%
35Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs68.8%
36Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs68.1%
37GPT 4.1· Stanford HELM specialized suites data docs67.4%
38GPT-4.1 mini· Stanford HELM specialized suites data docs65.2%
39Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs65.2%
40GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs64.5%
41GPT-5 mini· Stanford HELM specialized suites data docs64.5%
42Qwen2 72B Instruct· Stanford HELM specialized suites data docs64.5%
43O1 Mini· Stanford HELM specialized suites data docs63.8%
44DeepSeek V3· Stanford HELM specialized suites data docs63%
45GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs63%
46GPT-4o mini· Stanford HELM specialized suites data docs63%
47Claude 3 Opus· Stanford HELM specialized suites data docs62.3%
48GPT-4.1 nano· Stanford HELM specialized suites data docs62.3%
49Mistral Small· Stanford HELM specialized suites data docs62.3%
50DBRX-instruct· Stanford HELM specialized suites data docs60.9%

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