Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #13: Harassment

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
0.87

Score distribution

63.4%
Mean
64%
Median
21.7%
Std dev
32.3%
10th percentile
91.4%
90th percentile
13.4%
Min
13.4%98.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.87 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 docs98.7%
2Claude Haiku 4.5· Stanford HELM specialized suites data docs97.8%
3GPT OSS 20b· Stanford HELM specialized suites data docs96.2%
4GPT OSS 120B· Stanford HELM specialized suites data docs95.4%
5Claude 3 Haiku· Stanford HELM specialized suites data docs94.1%
6Claude Sonnet 4.5· Stanford HELM specialized suites data docs93.8%
7Claude Sonnet 4· Stanford HELM specialized suites data docs91.9%
8Claude 3 Sonnet· Stanford HELM specialized suites data docs91.4%
9Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs90.3%
10GPT 5.1· Stanford HELM specialized suites data docs89.8%
11Claude Opus 4· Stanford HELM specialized suites data docs89.2%
12Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs88.7%
13GPT-5 nano· Stanford HELM specialized suites data docs87.4%
14Palmyra X5· Stanford HELM specialized suites data docs85.2%
15Claude 3 Opus· Stanford HELM specialized suites data docs84.7%
16Granite 4.0 H Small· Stanford HELM specialized suites data docs83.9%
17Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs83.6%
18GPT-4.5 Preview· Stanford HELM specialized suites data docs83.1%
19Claude Sonnet 3.7· Stanford HELM specialized suites data docs82.5%
20GPT 5· Stanford HELM specialized suites data docs81.7%
21Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs81.5%
22Kimi K2 Instruct· Stanford HELM specialized suites data docs80.9%
23GPT-5 mini· Stanford HELM specialized suites data docs79.8%
24GPT-4 Turbo· Stanford HELM specialized suites data docs76.6%
25Llama3 8B Chat· Stanford HELM specialized suites data docs73.9%
26o1· Stanford HELM specialized suites data docs73.7%
27GPT-4.1 nano· Stanford HELM specialized suites data docs72.8%
28Qwen2 72B Instruct· Stanford HELM specialized suites data docs72.3%
29GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs72%
30o3 Mini· Stanford HELM specialized suites data docs71%
31OpenAI o3· Stanford HELM specialized suites data docs70.7%
32Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs70.2%
33GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs69.9%
34Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs69.6%
35Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs66.9%
36Gemini 3 Pro Preview· Stanford HELM specialized suites data docs66.1%
37o4 Mini· Stanford HELM specialized suites data docs65.9%
38GPT 4.1· Stanford HELM specialized suites data docs64%
39Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs64%
40Llama 4 Maverick· Stanford HELM specialized suites data docs63.7%
41Gemini 2.0 Flash· Stanford HELM specialized suites data docs63.2%
42gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs63.2%
43GPT-4.1 mini· Stanford HELM specialized suites data docs62.4%
44GPT 4osetting not named by the source· Stanford HELM specialized suites data docs59.4%
45GPT 4· Stanford HELM specialized suites data docs59.1%
46Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs59.1%
47gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs56.5%
48Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs55.6%
49GLM 4.5 Air· Stanford HELM specialized suites data docs55.1%
50GPT 4osetting not named by the source· Stanford HELM specialized suites data docs54.8%

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