Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #11: Weapon Usage & Development

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
1.28

Score distribution

68.3%
Mean
75.6%
Median
24.8%
Std dev
31.1%
10th percentile
95.6%
90th percentile
5.6%
Min
5.6%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.28 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
1OpenAI o3· Stanford HELM specialized suites data docs100%
2Claude 3 Haiku· Stanford HELM specialized suites data docs98.9%
3GPT OSS 120B· Stanford HELM specialized suites data docs98.9%
4Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs97.2%
5Claude 3 Opus· Stanford HELM specialized suites data docs96.7%
6Claude 3 Sonnet· Stanford HELM specialized suites data docs96.7%
7o1· Stanford HELM specialized suites data docs96.7%
8o4 Mini· Stanford HELM specialized suites data docs95.6%
9GPT OSS 20b· Stanford HELM specialized suites data docs94.4%
10GPT 5· Stanford HELM specialized suites data docs93.3%
11GPT-5 mini· Stanford HELM specialized suites data docs93.3%
12GPT-5 nano· Stanford HELM specialized suites data docs93.3%
13Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs93.3%
14Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs91.1%
15GPT 5.1· Stanford HELM specialized suites data docs91.1%
16Claude Haiku 4.5· Stanford HELM specialized suites data docs90%
17o3 Mini· Stanford HELM specialized suites data docs90%
18Palmyra X5· Stanford HELM specialized suites data docs88.9%
19Llama3 8B Chat· Stanford HELM specialized suites data docs85.6%
20Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs85.6%
21Kimi K2 Instruct· Stanford HELM specialized suites data docs84.4%
22Claude Sonnet 4.5· Stanford HELM specialized suites data docs83.9%
23Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs83.3%
24Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs83.3%
25GPT-4.5 Preview· Stanford HELM specialized suites data docs82.2%
26Granite 4.0 H Small· Stanford HELM specialized suites data docs82.2%
27Llama 4 Maverick· Stanford HELM specialized suites data docs82.2%
28Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs81.1%
29Gemini 3 Pro Preview· Stanford HELM specialized suites data docs81.1%
30GPT-4.1 nano· Stanford HELM specialized suites data docs81.1%
31GPT-4o mini· Stanford HELM specialized suites data docs81.1%
32GPT 4· Stanford HELM specialized suites data docs80%
33gemini-1.5-pro-002· Stanford HELM specialized suites data docs78.9%
34Claude Opus 4· Stanford HELM specialized suites data docs78.3%
35Claude Sonnet 4· Stanford HELM specialized suites data docs78.3%
36Gemini 2.0 Flash· Stanford HELM specialized suites data docs76.7%
37GPT-4 Turbo· Stanford HELM specialized suites data docs76.7%
38Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs75.6%
39GPT 4.1· Stanford HELM specialized suites data docs75.6%
40Claude Sonnet 3.7· Stanford HELM specialized suites data docs75%
41GPT 4osetting not named by the source· Stanford HELM specialized suites data docs74.4%
42GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs72.8%
43Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs72.2%
44Qwen2 72B Instruct· Stanford HELM specialized suites data docs72.2%
45GPT-4.1 mini· Stanford HELM specialized suites data docs71.7%
46gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs70.6%
47gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs70%
48gemini-1.5-flash-002· Stanford HELM specialized suites data docs69.4%
49GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs67.8%
50Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs62.2%

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