Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #36: Misrepresentation

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
1.19

Score distribution

56%
Mean
48.9%
Median
25.6%
Std dev
26.7%
10th percentile
91.1%
90th percentile
10%
Min
10%98.9%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.19 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
1GPT 5· Stanford HELM specialized suites data docs98.9%
2GPT-5 nano· Stanford HELM specialized suites data docs98.9%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs94.4%
4GPT-5 mini· Stanford HELM specialized suites data docs94.4%
5GPT OSS 20b· Stanford HELM specialized suites data docs93.3%
6Claude Haiku 4.5· Stanford HELM specialized suites data docs92.2%
7Claude 3 Haiku· Stanford HELM specialized suites data docs91.1%
8Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs91.1%
9Claude Sonnet 4.5· Stanford HELM specialized suites data docs91.1%
10Gemini 3 Pro Preview· Stanford HELM specialized suites data docs91.1%
11GPT 5.1· Stanford HELM specialized suites data docs91.1%
12GPT OSS 120B· Stanford HELM specialized suites data docs91.1%
13o1· Stanford HELM specialized suites data docs91.1%
14OpenAI o3· Stanford HELM specialized suites data docs91.1%
15Claude Sonnet 4· Stanford HELM specialized suites data docs90%
16o4 Mini· Stanford HELM specialized suites data docs87.8%
17Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs86.7%
18Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs86.7%
19Claude Opus 4· Stanford HELM specialized suites data docs85.6%
20Claude 3 Opus· Stanford HELM specialized suites data docs83.3%
21Kimi K2 Instruct· Stanford HELM specialized suites data docs83.3%
22Claude Sonnet 3.7· Stanford HELM specialized suites data docs82.2%
23Claude 3 Sonnet· Stanford HELM specialized suites data docs80%
24Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs80%
25o3 Mini· Stanford HELM specialized suites data docs67.8%
26Palmyra X5· Stanford HELM specialized suites data docs64.4%
27Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs61.1%
28GPT-4.5 Preview· Stanford HELM specialized suites data docs60%
29gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs58.9%
30GPT-4 Turbo· Stanford HELM specialized suites data docs54.4%
31gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs52.2%
32Llama 4 Maverick· Stanford HELM specialized suites data docs52.2%
33Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs51.1%
34Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs51.1%
35Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs50%
36Granite 4.0 H Small· Stanford HELM specialized suites data docs50%
37Llama3 8B Chat· Stanford HELM specialized suites data docs50%
38Gemini 2.0 Flash· Stanford HELM specialized suites data docs48.9%
39DeepSeek R1· Stanford HELM specialized suites data docs47.8%
40GPT 4· Stanford HELM specialized suites data docs46.7%
41GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs46.7%
42GLM 4.5 Air· Stanford HELM specialized suites data docs45.6%
43GPT 4.1· Stanford HELM specialized suites data docs45.6%
44GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs45.6%
45GPT-4.1 mini· Stanford HELM specialized suites data docs45.6%
46GPT-4.1 nano· Stanford HELM specialized suites data docs45.6%
47Grok 3 Mini· Stanford HELM specialized suites data docs45.6%
48Grok 3· Stanford HELM specialized suites data docs44.4%
49Qwen2 72B Instruct· Stanford HELM specialized suites data docs44.4%
50gemini-1.5-pro-002· Stanford HELM specialized suites data docs43.3%

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