Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Political Usage

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
1.01

Score distribution

62.9%
Mean
67.1%
Median
26.3%
Std dev
21.1%
10th percentile
94%
90th percentile
4.5%
Min
4.5%97.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.01 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 docs97.3%
2Claude Haiku 4.5· Stanford HELM specialized suites data docs97.3%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs97.2%
4GPT OSS 20b· Stanford HELM specialized suites data docs96.8%
5Claude 3 Haiku· Stanford HELM specialized suites data docs95.9%
6GPT OSS 120B· Stanford HELM specialized suites data docs95.7%
7Claude 3 Opus· Stanford HELM specialized suites data docs94.3%
8OpenAI o3· Stanford HELM specialized suites data docs94%
9Claude Sonnet 4.5· Stanford HELM specialized suites data docs92.8%
10GPT-5 mini· Stanford HELM specialized suites data docs92.4%
11Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs91.3%
12Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs90.7%
13o4 Mini· Stanford HELM specialized suites data docs90.1%
14GPT-5 nano· Stanford HELM specialized suites data docs89.1%
15o1· Stanford HELM specialized suites data docs89.1%
16Claude Sonnet 4· Stanford HELM specialized suites data docs88.9%
17GPT 5· Stanford HELM specialized suites data docs88.9%
18Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs87.6%
19Palmyra X5· Stanford HELM specialized suites data docs87.6%
20GPT 5.1· Stanford HELM specialized suites data docs87.5%
21Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs87.3%
22Claude Opus 4· Stanford HELM specialized suites data docs84%
23Kimi K2 Instruct· Stanford HELM specialized suites data docs83.6%
24o3 Mini· Stanford HELM specialized suites data docs82.8%
25Llama3 8B Chat· Stanford HELM specialized suites data docs82.5%
26Claude Sonnet 3.7· Stanford HELM specialized suites data docs82.1%
27GPT-4.5 Preview· Stanford HELM specialized suites data docs77.7%
28Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs76%
29Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs74.9%
30Llama 4 Maverick· Stanford HELM specialized suites data docs73.7%
31Granite 4.0 H Small· Stanford HELM specialized suites data docs72.7%
32Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs71.3%
33gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs69.3%
34GPT-4 Turbo· Stanford HELM specialized suites data docs69.1%
35Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs68.9%
36Qwen2 72B Instruct· Stanford HELM specialized suites data docs68.4%
37Gemini 2.0 Flash· Stanford HELM specialized suites data docs67.5%
38GPT-4.1 nano· Stanford HELM specialized suites data docs67.1%
39gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs66.4%
40Gemini 3 Pro Preview· Stanford HELM specialized suites data docs65.5%
41GPT 4.1· Stanford HELM specialized suites data docs64.9%
42gemini-1.5-flash-002· Stanford HELM specialized suites data docs64.1%
43Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs63.9%
44gemini-1.5-pro-002· Stanford HELM specialized suites data docs62%
45GPT-4.1 mini· Stanford HELM specialized suites data docs61.2%
46GLM 4.5 Air· Stanford HELM specialized suites data docs60.3%
47Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs58.1%
48GPT 4· Stanford HELM specialized suites data docs57.5%
49DeepSeek R1· Stanford HELM specialized suites data docs56.9%
50GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs55.9%

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