Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #43: Illegal/Regulated substances/goods

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
1.68

Score distribution

85.9%
Mean
93.3%
Median
21.4%
Std dev
46.7%
10th percentile
100%
90th percentile
13.3%
Min
13.3%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.68 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
51GPT 4osetting not named by the source· Stanford HELM specialized suites data docs86.7%
52GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs86.7%
53GPT-4.1 mini· Stanford HELM specialized suites data docs86.7%
54Kimi K2 Instruct· Stanford HELM specialized suites data docs86.7%
55Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs86.7%
56Yi 34B Chat· Stanford HELM specialized suites data docs86.7%
57DeepSeek V3· Stanford HELM specialized suites data docs80%
58Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs80%
59Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs80%
60Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs80%
61DeepSeek R1· Stanford HELM specialized suites data docs73.3%
62Grok 3· Stanford HELM specialized suites data docs73.3%
63Grok 3 Mini· Stanford HELM specialized suites data docs73.3%
64Grok 4· Stanford HELM specialized suites data docs66.7%
65Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs66.7%
66GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs60%
67Command R+· Stanford HELM specialized suites data docs53.3%
68Mistral Large2407· Stanford HELM specialized suites data docs46.7%
69Mistral Small· Stanford HELM specialized suites data docs40%
70Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs40%
71Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs40%
72O1 Mini· Stanford HELM specialized suites data docs33.3%
73Command R Plus· Stanford HELM specialized suites data docs26.7%
74DBRX-instruct· Stanford HELM specialized suites data docs20%
75Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs13.3%

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