Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #27: Disrupting Social Order (China-unique)

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
1.46

Score distribution

71.6%
Mean
81.4%
Median
27.2%
Std dev
25.6%
10th percentile
98.1%
90th percentile
3.1%
Min
3.1%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.46 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
1Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs100%
2Claude 3 Sonnet· Stanford HELM specialized suites data docs99.4%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs99.4%
4Claude Sonnet 4.5· Stanford HELM specialized suites data docs99.4%
5Claude Haiku 4.5· Stanford HELM specialized suites data docs98.9%
6GPT OSS 20b· Stanford HELM specialized suites data docs98.9%
7Claude 3 Haiku· Stanford HELM specialized suites data docs98.6%
8Claude Sonnet 4· Stanford HELM specialized suites data docs98.1%
9Claude 3 Opus· Stanford HELM specialized suites data docs97.8%
10GPT-5 mini· Stanford HELM specialized suites data docs97.8%
11GPT OSS 120B· Stanford HELM specialized suites data docs97.5%
12Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs96.9%
13Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs96.7%
14GPT-5 nano· Stanford HELM specialized suites data docs96.1%
15OpenAI o3· Stanford HELM specialized suites data docs96.1%
16GPT 5· Stanford HELM specialized suites data docs95.8%
17Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs95%
18Kimi K2 Instruct· Stanford HELM specialized suites data docs94.7%
19GPT 5.1· Stanford HELM specialized suites data docs94.4%
20Palmyra X5· Stanford HELM specialized suites data docs94.4%
21Claude Opus 4· Stanford HELM specialized suites data docs94.2%
22o4 Mini· Stanford HELM specialized suites data docs93.9%
23Claude Sonnet 3.7· Stanford HELM specialized suites data docs92.8%
24Llama3 8B Chat· Stanford HELM specialized suites data docs92.5%
25o1· Stanford HELM specialized suites data docs91.4%
26GPT-4.5 Preview· Stanford HELM specialized suites data docs88.9%
27Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs88.6%
28Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs88.1%
29Granite 4.0 H Small· Stanford HELM specialized suites data docs86.7%
30o3 Mini· Stanford HELM specialized suites data docs85.8%
31Qwen2 72B Instruct· Stanford HELM specialized suites data docs85.3%
32gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs83.9%
33GPT-4 Turbo· Stanford HELM specialized suites data docs83.1%
34Llama 4 Maverick· Stanford HELM specialized suites data docs83.1%
35Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs82.5%
36Gemini 3 Pro Preview· Stanford HELM specialized suites data docs81.7%
37gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs81.7%
38Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs81.4%
39GPT-4.1 nano· Stanford HELM specialized suites data docs81.4%
40DeepSeek R1· Stanford HELM specialized suites data docs80.3%
41Gemini 2.0 Flash· Stanford HELM specialized suites data docs78.6%
42GPT 4.1· Stanford HELM specialized suites data docs78.3%
43GLM 4.5 Air· Stanford HELM specialized suites data docs77.2%
44GPT-4.1 mini· Stanford HELM specialized suites data docs75%
45Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs74.4%
46gemini-1.5-flash-002· Stanford HELM specialized suites data docs74.2%
47gemini-1.5-pro-002· Stanford HELM specialized suites data docs73.6%
48GPT 4· Stanford HELM specialized suites data docs72.2%
49Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs71.4%
50GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs67.2%

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