Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #30: Disempowering Workers

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
1.03

Score distribution

64.2%
Mean
66.7%
Median
27%
Std dev
26.7%
10th percentile
96.7%
90th percentile
7.5%
Min
7.5%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.03 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· Stanford HELM specialized suites data docs100%
4Claude Sonnet 4.5· Stanford HELM specialized suites data docs99.2%
5Claude 3 Sonnet· Stanford HELM specialized suites data docs98.3%
6GPT 5.1· Stanford HELM specialized suites data docs98.3%
7Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs98.3%
8Claude Opus 4· Stanford HELM specialized suites data docs96.7%
9GPT OSS 120B· Stanford HELM specialized suites data docs96.7%
10GPT OSS 20b· Stanford HELM specialized suites data docs96.7%
11GPT 5· Stanford HELM specialized suites data docs95.8%
12Claude Sonnet 3.7· Stanford HELM specialized suites data docs95%
13Palmyra X5· Stanford HELM specialized suites data docs94.2%
14Claude 3 Haiku· Stanford HELM specialized suites data docs93.3%
15Claude 3 Opus· Stanford HELM specialized suites data docs93.3%
16Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs93.3%
17GPT-5 nano· Stanford HELM specialized suites data docs91.7%
18Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs90.8%
19OpenAI o3· Stanford HELM specialized suites data docs90%
20Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs90%
21Kimi K2 Instruct· Stanford HELM specialized suites data docs88.3%
22Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs87.5%
23GPT-5 mini· Stanford HELM specialized suites data docs87.5%
24o4 Mini· Stanford HELM specialized suites data docs86.7%
25GPT-4.5 Preview· Stanford HELM specialized suites data docs85%
26Llama3 8B Chat· Stanford HELM specialized suites data docs83.3%
27Gemini 3 Pro Preview· Stanford HELM specialized suites data docs80.8%
28o1· Stanford HELM specialized suites data docs80.8%
29gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs77.5%
30o3 Mini· Stanford HELM specialized suites data docs75.8%
31Gemini 2.0 Flash· Stanford HELM specialized suites data docs75%
32gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs74.2%
33gemini-1.5-flash-002· Stanford HELM specialized suites data docs73.3%
34gemini-1.5-pro-002· Stanford HELM specialized suites data docs73.3%
35Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs72.5%
36GLM 4.5 Air· Stanford HELM specialized suites data docs70.8%
37Llama 4 Maverick· Stanford HELM specialized suites data docs70.8%
38Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs66.7%
39Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs65%
40GPT 4.1· Stanford HELM specialized suites data docs63.3%
41GPT-4 Turbo· Stanford HELM specialized suites data docs63.3%
42GPT-4.1 nano· Stanford HELM specialized suites data docs63.3%
43Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs60%
44GPT-4.1 mini· Stanford HELM specialized suites data docs59.2%
45Granite 4.0 H Small· Stanford HELM specialized suites data docs59.2%
46GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs57.5%
47Qwen2 72B Instruct· Stanford HELM specialized suites data docs55.8%
48Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs52.5%
49Grok 3 Mini· Stanford HELM specialized suites data docs51.7%
50Llama 4 Scout· Stanford HELM specialized suites data docs50.8%

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