Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #34: Mis/disinformation

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
1.12

Score distribution

60.9%
Mean
66.7%
Median
26.4%
Std dev
20%
10th percentile
90%
90th percentile
6.7%
Min
6.7%96.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.12 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
1GPT OSS 20b· Stanford HELM specialized suites data docs96.7%
2Claude Haiku 4.5· Stanford HELM specialized suites data docs93.3%
3Claude Sonnet 4.5· Stanford HELM specialized suites data docs93.3%
4Claude 3 Opus· Stanford HELM specialized suites data docs90%
5Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs90%
6Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs90%
7Claude Opus 4· Stanford HELM specialized suites data docs90%
8GPT 5· Stanford HELM specialized suites data docs90%
9GPT OSS 120B· Stanford HELM specialized suites data docs90%
10GPT-5 mini· Stanford HELM specialized suites data docs90%
11GPT-5 nano· Stanford HELM specialized suites data docs90%
12Kimi K2 Instruct· Stanford HELM specialized suites data docs90%
13OpenAI o3· Stanford HELM specialized suites data docs90%
14Claude 3 Haiku· Stanford HELM specialized suites data docs86.7%
15Claude 3 Sonnet· Stanford HELM specialized suites data docs86.7%
16Claude Sonnet 3.7· Stanford HELM specialized suites data docs86.7%
17Claude Sonnet 4· Stanford HELM specialized suites data docs86.7%
18GPT 5.1· Stanford HELM specialized suites data docs86.7%
19GPT-4.5 Preview· Stanford HELM specialized suites data docs86.7%
20Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs86.7%
21o1· Stanford HELM specialized suites data docs86.7%
22o4 Mini· Stanford HELM specialized suites data docs86.7%
23Gemini 3 Pro Preview· Stanford HELM specialized suites data docs83.3%
24GPT 4.1· Stanford HELM specialized suites data docs80%
25Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs80%
26Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs80%
27Palmyra X5· Stanford HELM specialized suites data docs78.3%
28Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs73.3%
29Granite 4.0 H Small· Stanford HELM specialized suites data docs73.3%
30Llama 4 Maverick· Stanford HELM specialized suites data docs73.3%
31o3 Mini· Stanford HELM specialized suites data docs71.7%
32gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs70%
33GPT-4 Turbo· Stanford HELM specialized suites data docs70%
34GPT-4.1 mini· Stanford HELM specialized suites data docs70%
35Llama3 8B Chat· Stanford HELM specialized suites data docs70%
36gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs66.7%
37GLM 4.5 Air· Stanford HELM specialized suites data docs66.7%
38Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs66.7%
39Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs66.7%
40gemini-1.5-pro-002· Stanford HELM specialized suites data docs65%
41Gemini 2.0 Flash· Stanford HELM specialized suites data docs63.3%
42Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs63.3%
43Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs63.3%
44Qwen2 72B Instruct· Stanford HELM specialized suites data docs63.3%
45gemini-1.5-flash-002· Stanford HELM specialized suites data docs60%
46GPT 4osetting not named by the source· Stanford HELM specialized suites data docs60%
47GPT-4.1 nano· Stanford HELM specialized suites data docs60%
48Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs60%
49Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs56.7%
50GPT 4osetting not named by the source· Stanford HELM specialized suites data docs56.7%

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