Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #21: Endangerment, Harm, or Abuse of Children

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

51.3%
Mean
50.7%
Median
23.1%
Std dev
18.7%
10th percentile
82.7%
90th percentile
6.7%
Min
6.7%89.3%
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 Haiku 4.5· Stanford HELM specialized suites data docs89.3%
2Claude Sonnet 4· Stanford HELM specialized suites data docs86.7%
3Claude Sonnet 4.5· Stanford HELM specialized suites data docs86%
4Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs84.7%
5Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs84%
6Claude Opus 4· Stanford HELM specialized suites data docs84%
7Claude 3 Sonnet· Stanford HELM specialized suites data docs82.7%
8Palmyra X5· Stanford HELM specialized suites data docs82.7%
9Claude 3 Haiku· Stanford HELM specialized suites data docs81.3%
10GPT 5· Stanford HELM specialized suites data docs77.3%
11Llama 4 Maverick· Stanford HELM specialized suites data docs77.3%
12Claude 3 Opus· Stanford HELM specialized suites data docs76%
13Claude Sonnet 3.7· Stanford HELM specialized suites data docs76%
14GPT OSS 120B· Stanford HELM specialized suites data docs76%
15GPT OSS 20b· Stanford HELM specialized suites data docs76%
16GPT-5 nano· Stanford HELM specialized suites data docs76%
17GPT 5.1· Stanford HELM specialized suites data docs74.7%
18Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs74.7%
19OpenAI o3· Stanford HELM specialized suites data docs73.3%
20Gemini 3 Pro Preview· Stanford HELM specialized suites data docs72%
21o4 Mini· Stanford HELM specialized suites data docs72%
22GPT-4.5 Preview· Stanford HELM specialized suites data docs70.7%
23o3 Mini· Stanford HELM specialized suites data docs70%
24GPT-5 mini· Stanford HELM specialized suites data docs69.3%
25o1· Stanford HELM specialized suites data docs68%
26Kimi K2 Instruct· Stanford HELM specialized suites data docs65.3%
27Llama3 8B Chat· Stanford HELM specialized suites data docs64%
28Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs62.7%
29Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs62.7%
30Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs60%
31Granite 4.0 H Small· Stanford HELM specialized suites data docs58.7%
32gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs57.3%
33GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs54.7%
34gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs53.3%
35gemini-1.5-pro-002· Stanford HELM specialized suites data docs52%
36GPT-4 Turbo· Stanford HELM specialized suites data docs51.3%
37GPT 4.1· Stanford HELM specialized suites data docs50.7%
38GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs50.7%
39Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs50.7%
40Llama 4 Scout· Stanford HELM specialized suites data docs49.3%
41Gemini 2.0 Flash· Stanford HELM specialized suites data docs48.7%
42Grok 4· Stanford HELM specialized suites data docs48%
43Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs47.3%
44gemini-1.5-flash-002· Stanford HELM specialized suites data docs47.3%
45Qwen2 72B Instruct· Stanford HELM specialized suites data docs46.7%
46Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs45.3%
47GPT-4.1 mini· Stanford HELM specialized suites data docs44%
48GPT 4· Stanford HELM specialized suites data docs42.7%
49GLM 4.5 Air· Stanford HELM specialized suites data docs40%
50GPT-4.1 nano· Stanford HELM specialized suites data docs40%

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