Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Deception

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

60.8%
Mean
60.7%
Median
25.7%
Std dev
25.2%
10th percentile
93%
90th percentile
7.4%
Min
7.4%97.8%
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 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs97.8%
2Claude Haiku 4.5· Stanford HELM specialized suites data docs97%
3Claude Sonnet 4.5· Stanford HELM specialized suites data docs97%
4GPT OSS 20b· Stanford HELM specialized suites data docs95.6%
5GPT 5· Stanford HELM specialized suites data docs94.1%
6GPT OSS 120B· Stanford HELM specialized suites data docs94.1%
7GPT-5 nano· Stanford HELM specialized suites data docs93%
8OpenAI o3· Stanford HELM specialized suites data docs93%
9Claude Opus 4· Stanford HELM specialized suites data docs92.2%
10Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs91.9%
11Claude Sonnet 4· Stanford HELM specialized suites data docs91.9%
12GPT-5 mini· Stanford HELM specialized suites data docs91.9%
13Claude 3 Haiku· Stanford HELM specialized suites data docs91.1%
14GPT 5.1· Stanford HELM specialized suites data docs91.1%
15Claude 3 Sonnet· Stanford HELM specialized suites data docs90.4%
16Claude 3 Opus· Stanford HELM specialized suites data docs86.7%
17Kimi K2 Instruct· Stanford HELM specialized suites data docs86.7%
18Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs85.9%
19Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs85.9%
20o4 Mini· Stanford HELM specialized suites data docs83%
21Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs83%
22o1· Stanford HELM specialized suites data docs82.6%
23Gemini 3 Pro Preview· Stanford HELM specialized suites data docs81.5%
24Palmyra X5· Stanford HELM specialized suites data docs81.1%
25Claude Sonnet 3.7· Stanford HELM specialized suites data docs80.7%
26GPT-4.5 Preview· Stanford HELM specialized suites data docs80%
27Llama3 8B Chat· Stanford HELM specialized suites data docs74.4%
28o3 Mini· Stanford HELM specialized suites data docs74.4%
29Llama 4 Maverick· Stanford HELM specialized suites data docs73.7%
30Granite 4.0 H Small· Stanford HELM specialized suites data docs72.6%
31GPT 4.1· Stanford HELM specialized suites data docs71.1%
32Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs69.6%
33GPT-4 Turbo· Stanford HELM specialized suites data docs69.6%
34Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs69.3%
35gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs63.7%
36GPT-4.1 mini· Stanford HELM specialized suites data docs61.5%
37Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs61.5%
38Qwen2 72B Instruct· Stanford HELM specialized suites data docs60.7%
39Gemini 2.0 Flash· Stanford HELM specialized suites data docs59.6%
40Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs59.3%
41Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs55.6%
42GPT-4.1 nano· Stanford HELM specialized suites data docs55.6%
43gemini-1.5-pro-002· Stanford HELM specialized suites data docs55.2%
44GPT 4osetting not named by the source· Stanford HELM specialized suites data docs54.8%
45GPT 4· Stanford HELM specialized suites data docs54.4%
46GLM 4.5 Air· Stanford HELM specialized suites data docs54.1%
47gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs53.7%
48gemini-1.5-flash-002· Stanford HELM specialized suites data docs51.9%
49Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs50.7%
50Llama 4 Scout· Stanford HELM specialized suites data docs50%

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