Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #45: Other Illegal/Unlawful/Criminal Activities

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
1.88

Score distribution

82.1%
Mean
95%
Median
25.5%
Std dev
35%
10th percentile
100%
90th percentile
0%
Min
0%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.88 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 Sonnet· Stanford HELM specialized suites data docs100%
2Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs100%
3Claude Haiku 4.5· Stanford HELM specialized suites data docs100%
4GPT 5· Stanford HELM specialized suites data docs100%
5GPT 5.1· Stanford HELM specialized suites data docs100%
6GPT OSS 120B· Stanford HELM specialized suites data docs100%
7GPT OSS 20b· Stanford HELM specialized suites data docs100%
8GPT-5 mini· Stanford HELM specialized suites data docs100%
9GPT-5 nano· Stanford HELM specialized suites data docs100%
10Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs100%
11Llama3 8B Chat· Stanford HELM specialized suites data docs100%
12o4 Mini· Stanford HELM specialized suites data docs100%
13OpenAI o3· Stanford HELM specialized suites data docs100%
14Claude 3 Haiku· Stanford HELM specialized suites data docs98.3%
15Claude 3 Opus· Stanford HELM specialized suites data docs98.3%
16Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs98.3%
17Claude Sonnet 4· Stanford HELM specialized suites data docs98.3%
18Claude Sonnet 4.5· Stanford HELM specialized suites data docs98.3%
19Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs98.3%
20gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs98.3%
21GPT-4 Turbo· Stanford HELM specialized suites data docs98.3%
22GPT-4.5 Preview· Stanford HELM specialized suites data docs98.3%
23Kimi K2 Instruct· Stanford HELM specialized suites data docs98.3%
24o1· Stanford HELM specialized suites data docs98.3%
25o3 Mini· Stanford HELM specialized suites data docs98.3%
26Palmyra X5· Stanford HELM specialized suites data docs98.3%
27Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs98.3%
28Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs98.3%
29Claude Opus 4· Stanford HELM specialized suites data docs97.5%
30Claude Sonnet 3.7· Stanford HELM specialized suites data docs96.7%
31Gemini 2.0 Flash· Stanford HELM specialized suites data docs96.7%
32Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs96.7%
33GPT-4o mini· Stanford HELM specialized suites data docs96.7%
34Granite 4.0 H Small· Stanford HELM specialized suites data docs96.7%
35Llama 4 Maverick· Stanford HELM specialized suites data docs96.7%
36Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs95%
37GPT 4· Stanford HELM specialized suites data docs95%
38GPT-4.1 nano· Stanford HELM specialized suites data docs95%
39Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs95%
40GPT 4.1· Stanford HELM specialized suites data docs93.3%
41GPT 4osetting not named by the source· Stanford HELM specialized suites data docs93.3%
42GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs91.7%
43Qwen2 72B Instruct· Stanford HELM specialized suites data docs91.7%
44gemini-1.5-pro-002· Stanford HELM specialized suites data docs90%
45Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs90%
46Gemini 3 Pro Preview· Stanford HELM specialized suites data docs88.3%
47gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs88.3%
48GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs88.3%
49GPT-4.1 mini· Stanford HELM specialized suites data docs88.3%
50GLM 4.5 Air· Stanford HELM specialized suites data docs86.7%

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