Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - 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
1.73

Score distribution

78.8%
Mean
89.2%
Median
24.9%
Std dev
31.7%
10th percentile
99.2%
90th percentile
3.3%
Min
3.3%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.73 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 Haiku 4.5· Stanford HELM specialized suites data docs100%
3GPT OSS 20b· Stanford HELM specialized suites data docs100%
4Llama3 8B Chat· Stanford HELM specialized suites data docs100%
5Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs99.6%
6Claude 3 Haiku· Stanford HELM specialized suites data docs99.2%
7Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs99.2%
8Claude Sonnet 4.5· Stanford HELM specialized suites data docs99.2%
9GPT 5· Stanford HELM specialized suites data docs99.2%
10GPT OSS 120B· Stanford HELM specialized suites data docs99.2%
11GPT-5 mini· Stanford HELM specialized suites data docs99.2%
12Claude 3 Opus· Stanford HELM specialized suites data docs98.3%
13GPT-5 nano· Stanford HELM specialized suites data docs98.3%
14o4 Mini· Stanford HELM specialized suites data docs98.3%
15Claude Sonnet 4· Stanford HELM specialized suites data docs97.5%
16Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs97.5%
17Palmyra X5· Stanford HELM specialized suites data docs97.5%
18Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs97.5%
19GPT 5.1· Stanford HELM specialized suites data docs96.7%
20GPT-4.5 Preview· Stanford HELM specialized suites data docs96.7%
21Granite 4.0 H Small· Stanford HELM specialized suites data docs96.7%
22o3 Mini· Stanford HELM specialized suites data docs96.7%
23OpenAI o3· Stanford HELM specialized suites data docs96.7%
24Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs96.7%
25o1· Stanford HELM specialized suites data docs95.8%
26Claude Sonnet 3.7· Stanford HELM specialized suites data docs95%
27Llama 4 Maverick· Stanford HELM specialized suites data docs95%
28Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs94.2%
29Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs94.2%
30gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs94.2%
31GPT-4 Turbo· Stanford HELM specialized suites data docs94.2%
32Claude Opus 4· Stanford HELM specialized suites data docs93.3%
33Gemini 2.0 Flash· Stanford HELM specialized suites data docs92.5%
34Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs92.5%
35Kimi K2 Instruct· Stanford HELM specialized suites data docs92.5%
36Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs91.7%
37GPT-4.1 nano· Stanford HELM specialized suites data docs90%
38GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs89.2%
39GPT 4· Stanford HELM specialized suites data docs87.9%
40Gemini 3 Pro Preview· Stanford HELM specialized suites data docs87.5%
41GPT-4o mini· Stanford HELM specialized suites data docs87.5%
42GPT 4.1· Stanford HELM specialized suites data docs84.2%
43GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs84.2%
44Qwen2 72B Instruct· Stanford HELM specialized suites data docs84.2%
45gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs82.5%
46GPT 4osetting not named by the source· Stanford HELM specialized suites data docs82.5%
47gemini-1.5-pro-002· Stanford HELM specialized suites data docs80.8%
48Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs80.8%
49gemini-1.5-flash-002· Stanford HELM specialized suites data docs80%
50GPT-4.1 mini· Stanford HELM specialized suites data docs80%

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