Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Economic Harm

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
0.97

Score distribution

66.2%
Mean
67.3%
Median
26.6%
Std dev
26.3%
10th percentile
97.3%
90th percentile
6.3%
Min
6.3%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.97 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 docs99%
4Claude 3 Sonnet· Stanford HELM specialized suites data docs98.7%
5Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs98.3%
6Claude Sonnet 4· Stanford HELM specialized suites data docs98%
7GPT OSS 120B· Stanford HELM specialized suites data docs97.3%
8GPT OSS 20b· Stanford HELM specialized suites data docs97.3%
9Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs96.7%
10GPT 5.1· Stanford HELM specialized suites data docs96.7%
11Claude 3 Haiku· Stanford HELM specialized suites data docs96%
12GPT 5· Stanford HELM specialized suites data docs95.7%
13Claude Opus 4· Stanford HELM specialized suites data docs95.3%
14GPT-5 nano· Stanford HELM specialized suites data docs95.3%
15Claude 3 Opus· Stanford HELM specialized suites data docs95%
16Palmyra X5· Stanford HELM specialized suites data docs95%
17OpenAI o3· Stanford HELM specialized suites data docs94.7%
18GPT-5 mini· Stanford HELM specialized suites data docs93.7%
19Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs92.3%
20Claude Sonnet 3.7· Stanford HELM specialized suites data docs91.7%
21Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs91.7%
22o1· Stanford HELM specialized suites data docs89.7%
23Kimi K2 Instruct· Stanford HELM specialized suites data docs88.7%
24o4 Mini· Stanford HELM specialized suites data docs88.7%
25GPT-4.5 Preview· Stanford HELM specialized suites data docs88%
26Llama3 8B Chat· Stanford HELM specialized suites data docs87.3%
27Gemini 3 Pro Preview· Stanford HELM specialized suites data docs81.3%
28Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs79.7%
29o3 Mini· Stanford HELM specialized suites data docs79.7%
30Llama 4 Maverick· Stanford HELM specialized suites data docs79.3%
31gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs77.7%
32Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs75%
33Gemini 2.0 Flash· Stanford HELM specialized suites data docs73.7%
34Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs72.7%
35GPT-4 Turbo· Stanford HELM specialized suites data docs72.7%
36gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs72.3%
37gemini-1.5-pro-002· Stanford HELM specialized suites data docs70%
38Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs67.3%
39gemini-1.5-flash-002· Stanford HELM specialized suites data docs67.3%
40GLM 4.5 Air· Stanford HELM specialized suites data docs67.3%
41GPT-4.1 nano· Stanford HELM specialized suites data docs67.3%
42Granite 4.0 H Small· Stanford HELM specialized suites data docs67.3%
43GPT 4.1· Stanford HELM specialized suites data docs66.3%
44GPT-4.1 mini· Stanford HELM specialized suites data docs62.7%
45GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs62.3%
46Qwen2 72B Instruct· Stanford HELM specialized suites data docs61%
47Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs60%
48GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs57.3%
49Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs57%
50GPT-4o mini· Stanford HELM specialized suites data docs54.3%

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