Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #8: Celebrating Suffering

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
1.08

Score distribution

53.5%
Mean
51.4%
Median
15.5%
Std dev
36.2%
10th percentile
74.6%
90th percentile
30.4%
Min
30.4%94.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.08 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 Sonnet 4.5· Stanford HELM specialized suites data docs94.2%
2Claude Haiku 4.5· Stanford HELM specialized suites data docs92.8%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs86.2%
4Claude Sonnet 4· Stanford HELM specialized suites data docs84.8%
5GPT OSS 120B· Stanford HELM specialized suites data docs82.6%
6GPT OSS 20b· Stanford HELM specialized suites data docs79.7%
7Claude Opus 4· Stanford HELM specialized suites data docs79%
8GPT-5 mini· Stanford HELM specialized suites data docs74.6%
9GPT 5· Stanford HELM specialized suites data docs73.9%
10GPT 5.1· Stanford HELM specialized suites data docs73.9%
11Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs73.2%
12GPT-5 nano· Stanford HELM specialized suites data docs71%
13Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs69.6%
14Palmyra X5· Stanford HELM specialized suites data docs67.4%
15Claude Sonnet 3.7· Stanford HELM specialized suites data docs65.9%
16Claude 3 Opus· Stanford HELM specialized suites data docs64.5%
17Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs63.8%
18Claude 3 Sonnet· Stanford HELM specialized suites data docs61.6%
19Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs61.6%
20gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs61.6%
21Llama 4 Maverick· Stanford HELM specialized suites data docs61.6%
22OpenAI o3· Stanford HELM specialized suites data docs61.6%
23Claude 3 Haiku· Stanford HELM specialized suites data docs60.9%
24Kimi K2 Instruct· Stanford HELM specialized suites data docs60.9%
25Gemini 2.0 Flash· Stanford HELM specialized suites data docs59.4%
26Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs59.4%
27gemini-1.5-pro-002· Stanford HELM specialized suites data docs55.8%
28GLM 4.5 Air· Stanford HELM specialized suites data docs55.8%
29Granite 4.0 H Small· Stanford HELM specialized suites data docs55.8%
30Llama3 8B Chat· Stanford HELM specialized suites data docs55.8%
31DeepSeek R1· Stanford HELM specialized suites data docs55.1%
32gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs55.1%
33o4 Mini· Stanford HELM specialized suites data docs55.1%
34Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs53.6%
35o1· Stanford HELM specialized suites data docs52.9%
36o3 Mini· Stanford HELM specialized suites data docs52.2%
37Qwen2 72B Instruct· Stanford HELM specialized suites data docs52.2%
38gemini-1.5-flash-002· Stanford HELM specialized suites data docs51.4%
39GPT 4osetting not named by the source· Stanford HELM specialized suites data docs51.4%
40GPT-4.1 mini· Stanford HELM specialized suites data docs51.4%
41GPT 4.1· Stanford HELM specialized suites data docs50.7%
42Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs50.7%
43Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs50%
44GPT-4.5 Preview· Stanford HELM specialized suites data docs50%
45GPT-4 Turbo· Stanford HELM specialized suites data docs47.8%
46Grok 3 Mini· Stanford HELM specialized suites data docs47.8%
47Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs47.8%
48Gemini 3 Pro Preview· Stanford HELM specialized suites data docs44.2%
49GPT 4osetting not named by the source· Stanford HELM specialized suites data docs44.2%
50Llama 4 Scout· Stanford HELM specialized suites data docs44.2%

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