Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #24: Political Persuasion

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
0.69

Score distribution

46.1%
Mean
44.3%
Median
28.6%
Std dev
7.6%
10th percentile
89.5%
90th percentile
4.3%
Min
4.3%93.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.69 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 docs93.3%
2GPT OSS 120B· Stanford HELM specialized suites data docs93.3%
3GPT OSS 20b· Stanford HELM specialized suites data docs93.3%
4o1· Stanford HELM specialized suites data docs92.9%
5Claude Haiku 4.5· Stanford HELM specialized suites data docs92.4%
6OpenAI o3· Stanford HELM specialized suites data docs92.4%
7Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs91%
8Claude 3 Haiku· Stanford HELM specialized suites data docs89.5%
9o4 Mini· Stanford HELM specialized suites data docs88.1%
10Claude 3 Opus· Stanford HELM specialized suites data docs85.7%
11GPT-5 mini· Stanford HELM specialized suites data docs85.2%
12o3 Mini· Stanford HELM specialized suites data docs79%
13GPT-5 nano· Stanford HELM specialized suites data docs78.1%
14Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs77.6%
15Claude Sonnet 4.5· Stanford HELM specialized suites data docs77.1%
16GPT 5.1· Stanford HELM specialized suites data docs77.1%
17GPT 5· Stanford HELM specialized suites data docs74.8%
18Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs74.3%
19Palmyra X5· Stanford HELM specialized suites data docs74.3%
20Llama3 8B Chat· Stanford HELM specialized suites data docs73.3%
21Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs70.5%
22Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs69.5%
23Claude Sonnet 4· Stanford HELM specialized suites data docs67.1%
24Kimi K2 Instruct· Stanford HELM specialized suites data docs61.9%
25GPT-4.5 Preview· Stanford HELM specialized suites data docs61.4%
26Claude Opus 4· Stanford HELM specialized suites data docs59%
27Llama 4 Maverick· Stanford HELM specialized suites data docs57.6%
28Claude Sonnet 3.7· Stanford HELM specialized suites data docs56.2%
29Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs53.8%
30Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs53.3%
31gemini-1.5-flash-002· Stanford HELM specialized suites data docs52.4%
32Granite 4.0 H Small· Stanford HELM specialized suites data docs50%
33Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs49%
34Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs48.6%
35Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs47.6%
36gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs45.7%
37GPT-4 Turbo· Stanford HELM specialized suites data docs45.2%
38Gemini 2.0 Flash· Stanford HELM specialized suites data docs44.3%
39gemini-1.5-pro-002· Stanford HELM specialized suites data docs43.8%
40GPT-4.1 nano· Stanford HELM specialized suites data docs41%
41GPT 4.1· Stanford HELM specialized suites data docs40%
42Qwen2 72B Instruct· Stanford HELM specialized suites data docs39.5%
43Gemini 3 Pro Preview· Stanford HELM specialized suites data docs39%
44gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs36.2%
45GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs33.8%
46GPT-4.1 mini· Stanford HELM specialized suites data docs33.8%
47Llama 4 Scout· Stanford HELM specialized suites data docs33.3%
48GPT 4· Stanford HELM specialized suites data docs31.4%
49GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs31%
50GPT-4o mini· Stanford HELM specialized suites data docs30.5%

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