Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #6: Advice in Heavily Regulated Industries

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
0.17

Score distribution

10.9%
Mean
8.7%
Median
8.3%
Std dev
1.3%
10th percentile
22.7%
90th percentile
0%
Min
0%40.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.17 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
51gemini-1.5-pro-002· Stanford HELM specialized suites data docs6%
52GPT 5.1· Stanford HELM specialized suites data docs6%
53GPT-4.1 nano· Stanford HELM specialized suites data docs6%
54Llama 4 Maverick· Stanford HELM specialized suites data docs6%
55Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs6%
56DeepSeek R1· Stanford HELM specialized suites data docs5.3%
57Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs5.3%
58GLM 4.5 Air· Stanford HELM specialized suites data docs4.7%
59Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs4.7%
60GPT 4osetting not named by the source· Stanford HELM specialized suites data docs4%
61o4 Mini· Stanford HELM specialized suites data docs4%
62GPT-4 Turbo· Stanford HELM specialized suites data docs3.3%
63GPT 4osetting not named by the source· Stanford HELM specialized suites data docs2.7%
64GPT-4o mini· Stanford HELM specialized suites data docs2%
65Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs2%
66Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs2%
67Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs2%
68GPT-4.5 Preview· Stanford HELM specialized suites data docs1.3%
69Mistral Small· Stanford HELM specialized suites data docs1.3%
70Mistral Large2407· Stanford HELM specialized suites data docs0.667%
71Command R Plus· Stanford HELM specialized suites data docs0%
72DeepSeek V3· Stanford HELM specialized suites data docs0%
73GPT 4.1· Stanford HELM specialized suites data docs0%
74GPT-4.1 mini· Stanford HELM specialized suites data docs0%
75Kimi K2 Instruct· Stanford HELM specialized suites data docs0%

Showing 51–75 of 75 results.

Data credits: Stanford HELM (Stanford CRFM); scenario scores from Capabilities, Safety, Long Context, MedHELM, AIR-Bench, MMLU and Finance