Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #22: Child Sexual Abuse

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
1.40

Score distribution

74.1%
Mean
80%
Median
21.8%
Std dev
40%
10th percentile
96.7%
90th percentile
6.7%
Min
6.7%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.40 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 docs100%
2Claude Haiku 4.5· Stanford HELM specialized suites data docs100%
3Claude Sonnet 4.5· Stanford HELM specialized suites data docs100%
4GPT OSS 20b· Stanford HELM specialized suites data docs100%
5Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs100%
6Claude 3 Sonnet· Stanford HELM specialized suites data docs96.7%
7GPT OSS 120B· Stanford HELM specialized suites data docs96.7%
8Llama3 8B Chat· Stanford HELM specialized suites data docs96.7%
9Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs96.7%
10Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs95%
11Claude 3 Haiku· Stanford HELM specialized suites data docs93.3%
12Claude Opus 4· Stanford HELM specialized suites data docs93.3%
13Claude Sonnet 4· Stanford HELM specialized suites data docs93.3%
14GPT-4.5 Preview· Stanford HELM specialized suites data docs93.3%
15GPT-5 mini· Stanford HELM specialized suites data docs93.3%
16GPT-5 nano· Stanford HELM specialized suites data docs93.3%
17Llama 4 Maverick· Stanford HELM specialized suites data docs93.3%
18Palmyra X5· Stanford HELM specialized suites data docs93.3%
19Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs93.3%
20Claude 3 Opus· Stanford HELM specialized suites data docs91.7%
21Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs90%
22GPT 5.1· Stanford HELM specialized suites data docs90%
23o4 Mini· Stanford HELM specialized suites data docs90%
24o1· Stanford HELM specialized suites data docs88.3%
25GPT 5· Stanford HELM specialized suites data docs86.7%
26GPT-4 Turbo· Stanford HELM specialized suites data docs86.7%
27Granite 4.0 H Small· Stanford HELM specialized suites data docs86.7%
28Kimi K2 Instruct· Stanford HELM specialized suites data docs86.7%
29o3 Mini· Stanford HELM specialized suites data docs86.7%
30OpenAI o3· Stanford HELM specialized suites data docs86.7%
31GPT-4.1 nano· Stanford HELM specialized suites data docs85%
32Claude Sonnet 3.7· Stanford HELM specialized suites data docs83.3%
33gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs83.3%
34GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs83.3%
35Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs83.3%
36Gemini 2.0 Flash· Stanford HELM specialized suites data docs80%
37Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs80%
38Gemini 3 Pro Preview· Stanford HELM specialized suites data docs80%
39GPT 4· Stanford HELM specialized suites data docs80%
40Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs76.7%
41gemini-1.5-pro-002· Stanford HELM specialized suites data docs76.7%
42GPT 4.1· Stanford HELM specialized suites data docs76.7%
43GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs76.7%
44Llama 4 Scout· Stanford HELM specialized suites data docs76.7%
45gemini-1.5-flash-002· Stanford HELM specialized suites data docs73.3%
46Grok 3 Mini· Stanford HELM specialized suites data docs71.7%
47gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs70%
48GLM 4.5 Air· Stanford HELM specialized suites data docs70%
49GPT-4.1 mini· Stanford HELM specialized suites data docs70%
50GPT 4osetting not named by the source· Stanford HELM specialized suites data docs66.7%

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