Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #2: Integrity

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
1.59

Score distribution

78.9%
Mean
88.9%
Median
24.1%
Std dev
34.6%
10th percentile
100%
90th percentile
9.9%
Min
9.9%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.59 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 Haiku· Stanford HELM specialized suites data docs100%
2Claude 3 Sonnet· Stanford HELM specialized suites data docs100%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs100%
4Claude Haiku 4.5· Stanford HELM specialized suites data docs100%
5GPT 5· Stanford HELM specialized suites data docs100%
6GPT OSS 120B· Stanford HELM specialized suites data docs100%
7GPT OSS 20b· Stanford HELM specialized suites data docs100%
8GPT-5 mini· Stanford HELM specialized suites data docs100%
9GPT-5 nano· Stanford HELM specialized suites data docs100%
10o4 Mini· Stanford HELM specialized suites data docs100%
11OpenAI o3· Stanford HELM specialized suites data docs100%
12Claude 3 Opus· Stanford HELM specialized suites data docs98.8%
13Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs98.8%
14GPT 5.1· Stanford HELM specialized suites data docs98.8%
15o3 Mini· Stanford HELM specialized suites data docs98.8%
16Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs98.8%
17Llama3 8B Chat· Stanford HELM specialized suites data docs98.1%
18Claude Sonnet 4.5· Stanford HELM specialized suites data docs97.5%
19o1· Stanford HELM specialized suites data docs97.5%
20Palmyra X5· Stanford HELM specialized suites data docs97.5%
21Kimi K2 Instruct· Stanford HELM specialized suites data docs96.3%
22GPT 4· Stanford HELM specialized suites data docs95.7%
23GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs95.7%
24GPT-4o mini· Stanford HELM specialized suites data docs95.1%
25Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs95.1%
26Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs95.1%
27GPT-4.5 Preview· Stanford HELM specialized suites data docs94.4%
28GPT 4osetting not named by the source· Stanford HELM specialized suites data docs93.8%
29GPT-4 Turbo· Stanford HELM specialized suites data docs93.8%
30GPT-4.1 nano· Stanford HELM specialized suites data docs93.2%
31GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs92.6%
32Claude Opus 4· Stanford HELM specialized suites data docs92%
33Gemini 3 Pro Preview· Stanford HELM specialized suites data docs92%
34Claude Sonnet 4· Stanford HELM specialized suites data docs90.7%
35GPT 4.1· Stanford HELM specialized suites data docs90.1%
36Granite 4.0 H Small· Stanford HELM specialized suites data docs90.1%
37Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs89.5%
38Claude Sonnet 3.7· Stanford HELM specialized suites data docs88.9%
39Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs88.9%
40Gemini 2.0 Flash· Stanford HELM specialized suites data docs87.7%
41Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs87.7%
42gemini-1.5-flash-002· Stanford HELM specialized suites data docs86.4%
43Qwen2 72B Instruct· Stanford HELM specialized suites data docs86.4%
44Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs85.8%
45GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs84.6%
46GPT-4.1 mini· Stanford HELM specialized suites data docs84.6%
47gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs83.3%
48Llama 4 Maverick· Stanford HELM specialized suites data docs82.7%
49GPT 4osetting not named by the source· Stanford HELM specialized suites data docs79.6%
50gemini-1.5-pro-002· Stanford HELM specialized suites data docs76.5%

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