Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #41-42: Privacy Violations/Sensitive Data Combinations

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
1.06

Score distribution

69.3%
Mean
69.4%
Median
18.9%
Std dev
43.3%
10th percentile
91.6%
90th percentile
19.4%
Min
19.4%96.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.06 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 Haiku 4.5· Stanford HELM specialized suites data docs96.5%
2GPT OSS 120B· Stanford HELM specialized suites data docs94.8%
3Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs94.1%
4Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs93.4%
5Claude Sonnet 4.5· Stanford HELM specialized suites data docs92.8%
6GPT OSS 20b· Stanford HELM specialized suites data docs92.7%
7GPT-5 nano· Stanford HELM specialized suites data docs92.4%
8Claude 3 Sonnet· Stanford HELM specialized suites data docs91.6%
9Claude 3 Haiku· Stanford HELM specialized suites data docs91.6%
10Claude 3 Opus· Stanford HELM specialized suites data docs90.9%
11Claude Sonnet 4· Stanford HELM specialized suites data docs90.8%
12GPT 5· Stanford HELM specialized suites data docs90.2%
13GPT-5 mini· Stanford HELM specialized suites data docs90%
14Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs89.9%
15OpenAI o3· Stanford HELM specialized suites data docs88.9%
16Claude Opus 4· Stanford HELM specialized suites data docs88.1%
17GPT 5.1· Stanford HELM specialized suites data docs87.3%
18o1· Stanford HELM specialized suites data docs87.2%
19o4 Mini· Stanford HELM specialized suites data docs85.9%
20Claude Sonnet 3.7· Stanford HELM specialized suites data docs85.5%
21Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs84.7%
22Gemini 3 Pro Preview· Stanford HELM specialized suites data docs83.2%
23Palmyra X5· Stanford HELM specialized suites data docs82.7%
24Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs82.4%
25Kimi K2 Instruct· Stanford HELM specialized suites data docs81.6%
26o3 Mini· Stanford HELM specialized suites data docs81.4%
27GPT-4 Turbo· Stanford HELM specialized suites data docs80.8%
28GPT-4.5 Preview· Stanford HELM specialized suites data docs80.6%
29Granite 4.0 H Small· Stanford HELM specialized suites data docs80%
30Llama3 8B Chat· Stanford HELM specialized suites data docs78%
31Llama 4 Maverick· Stanford HELM specialized suites data docs76.3%
32GPT 4· Stanford HELM specialized suites data docs75.3%
33Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs74.9%
34GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs73.8%
35Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs73%
36GPT-4.1 nano· Stanford HELM specialized suites data docs71.8%
37GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs69.5%
38gemini-1.5-pro-002· Stanford HELM specialized suites data docs69.4%
39gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs69.4%
40GPT 4osetting not named by the source· Stanford HELM specialized suites data docs69.4%
41GPT 4.1· Stanford HELM specialized suites data docs69.3%
42gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs69.1%
43Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs68.8%
44Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs68.8%
45Gemini 2.0 Flash· Stanford HELM specialized suites data docs68%
46gemini-1.5-flash-002· Stanford HELM specialized suites data docs67.7%
47GPT-4o mini· Stanford HELM specialized suites data docs66.9%
48GPT-4.1 mini· Stanford HELM specialized suites data docs65.3%
49Qwen2 72B Instruct· Stanford HELM specialized suites data docs65.1%
50Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs64.3%

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