Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #1: Confidentiality

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
1.44

Score distribution

68.3%
Mean
75.2%
Median
27.5%
Std dev
24.8%
10th percentile
99%
90th percentile
1.4%
Min
1.4%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.44 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 Haiku· Stanford HELM specialized suites data docs100%
2Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs100%
3Claude Sonnet 4.5· Stanford HELM specialized suites data docs100%
4GPT OSS 120B· Stanford HELM specialized suites data docs100%
5Claude 3 Sonnet· Stanford HELM specialized suites data docs99%
6Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs99%
7Claude Haiku 4.5· Stanford HELM specialized suites data docs99%
8GPT 5· Stanford HELM specialized suites data docs99%
9OpenAI o3· Stanford HELM specialized suites data docs99%
10GPT 5.1· Stanford HELM specialized suites data docs98.1%
11GPT OSS 20b· Stanford HELM specialized suites data docs98.1%
12GPT-5 mini· Stanford HELM specialized suites data docs98.1%
13GPT-5 nano· Stanford HELM specialized suites data docs98.1%
14o4 Mini· Stanford HELM specialized suites data docs97.1%
15Claude Opus 4· Stanford HELM specialized suites data docs91.9%
16Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs91.4%
17Claude Sonnet 4· Stanford HELM specialized suites data docs91%
18Claude Sonnet 3.7· Stanford HELM specialized suites data docs90.5%
19o1· Stanford HELM specialized suites data docs90.5%
20Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs90%
21Llama3 8B Chat· Stanford HELM specialized suites data docs89.5%
22Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs89.5%
23Claude 3 Opus· Stanford HELM specialized suites data docs89%
24Granite 4.0 H Small· Stanford HELM specialized suites data docs87.1%
25Palmyra X5· Stanford HELM specialized suites data docs86.7%
26GPT 4· Stanford HELM specialized suites data docs85.7%
27GPT-4.5 Preview· Stanford HELM specialized suites data docs85.7%
28Kimi K2 Instruct· Stanford HELM specialized suites data docs85.7%
29o3 Mini· Stanford HELM specialized suites data docs85.7%
30Gemini 3 Pro Preview· Stanford HELM specialized suites data docs84.3%
31Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs83.8%
32GPT-4 Turbo· Stanford HELM specialized suites data docs80%
33GPT 4.1· Stanford HELM specialized suites data docs79%
34GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs79%
35GPT-4.1 nano· Stanford HELM specialized suites data docs79%
36Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs77.1%
37GPT 4osetting not named by the source· Stanford HELM specialized suites data docs76.2%
38Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs75.2%
39GPT-4o mini· Stanford HELM specialized suites data docs75.2%
40Gemini 2.0 Flash· Stanford HELM specialized suites data docs73.3%
41GPT-4.1 mini· Stanford HELM specialized suites data docs73.3%
42Llama 4 Maverick· Stanford HELM specialized suites data docs73.3%
43gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs71.9%
44GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs71.4%
45Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs66.7%
46Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs65.2%
47Qwen2 72B Instruct· Stanford HELM specialized suites data docs64.8%
48gemini-1.5-flash-002· Stanford HELM specialized suites data docs64.3%
49GPT 4osetting not named by the source· Stanford HELM specialized suites data docs63.8%
50gemini-1.5-pro-002· Stanford HELM specialized suites data docs62.9%

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