Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Security Risks

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
1.37

Score distribution

72.6%
Mean
81.3%
Median
25.6%
Std dev
31.5%
10th percentile
99.1%
90th percentile
4.7%
Min
4.7%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.37 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 docs100%
2GPT OSS 120B· Stanford HELM specialized suites data docs100%
3Claude 3 Haiku· Stanford HELM specialized suites data docs99.5%
4Claude Haiku 4.5· Stanford HELM specialized suites data docs99.5%
5Claude 3 Sonnet· Stanford HELM specialized suites data docs99.1%
6Claude Sonnet 4.5· Stanford HELM specialized suites data docs99.1%
7GPT OSS 20b· Stanford HELM specialized suites data docs99.1%
8GPT-5 mini· Stanford HELM specialized suites data docs99.1%
9GPT-5 nano· Stanford HELM specialized suites data docs99.1%
10OpenAI o3· Stanford HELM specialized suites data docs99.1%
11o4 Mini· Stanford HELM specialized suites data docs98.6%
12Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs98.4%
13GPT 5· Stanford HELM specialized suites data docs98.2%
14GPT 5.1· Stanford HELM specialized suites data docs97.3%
15Claude 3 Opus· Stanford HELM specialized suites data docs93.7%
16Llama3 8B Chat· Stanford HELM specialized suites data docs93.5%
17Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs93.5%
18o1· Stanford HELM specialized suites data docs93.2%
19Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs92.6%
20Claude Opus 4· Stanford HELM specialized suites data docs91.7%
21Claude Sonnet 4· Stanford HELM specialized suites data docs91.4%
22o3 Mini· Stanford HELM specialized suites data docs91%
23Kimi K2 Instruct· Stanford HELM specialized suites data docs90.5%
24Palmyra X5· Stanford HELM specialized suites data docs90.1%
25GPT 4· Stanford HELM specialized suites data docs89.9%
26Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs89.6%
27GPT-4.5 Preview· Stanford HELM specialized suites data docs88.5%
28Claude Sonnet 3.7· Stanford HELM specialized suites data docs87.6%
29Granite 4.0 H Small· Stanford HELM specialized suites data docs87.6%
30GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs86.7%
31Gemini 3 Pro Preview· Stanford HELM specialized suites data docs86%
32GPT-4 Turbo· Stanford HELM specialized suites data docs86%
33GPT-4o mini· Stanford HELM specialized suites data docs83.8%
34GPT-4.1 nano· Stanford HELM specialized suites data docs83.6%
35GPT 4osetting not named by the source· Stanford HELM specialized suites data docs83.3%
36Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs83.3%
37GPT 4.1· Stanford HELM specialized suites data docs82.9%
38Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs81.3%
39Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs81.1%
40Gemini 2.0 Flash· Stanford HELM specialized suites data docs79.7%
41GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs79.7%
42GPT-4.1 mini· Stanford HELM specialized suites data docs77.7%
43Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs76.8%
44gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs76.6%
45Llama 4 Maverick· Stanford HELM specialized suites data docs76.6%
46gemini-1.5-flash-002· Stanford HELM specialized suites data docs73.6%
47Qwen2 72B Instruct· Stanford HELM specialized suites data docs72.5%
48GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs72.3%
49GPT 4osetting not named by the source· Stanford HELM specialized suites data docs70.5%
50gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs69.8%

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