Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #9: Violent acts

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
1.86

Score distribution

77%
Mean
86.7%
Median
24.5%
Std dev
35%
10th percentile
98.3%
90th percentile
5%
Min
5%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.86 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-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs80%
52DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs70%
53Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs68.3%
54Yi 34B Chat· Stanford HELM specialized suites data docs66.7%
55GPT 4osetting not named by the source· Stanford HELM specialized suites data docs65%
56GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs65%
57Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs63.3%
58Grok 3 Mini· Stanford HELM specialized suites data docs60%
59Llama 4 Scout· Stanford HELM specialized suites data docs55%
60Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs55%
61DeepSeek R1· Stanford HELM specialized suites data docs52.5%
62GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs51.7%
63Grok 4· Stanford HELM specialized suites data docs51.7%
64Grok 3· Stanford HELM specialized suites data docs48.3%
65DeepSeek V3· Stanford HELM specialized suites data docs45%
66Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs41.7%
67O1 Mini· Stanford HELM specialized suites data docs40%
68Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs35%
69Command R+· Stanford HELM specialized suites data docs28.3%
70Mistral Large2407· Stanford HELM specialized suites data docs26.7%
71Command R Plus· Stanford HELM specialized suites data docs25%
72Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs25%
73Mistral Small· Stanford HELM specialized suites data docs23.3%
74Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs21.7%
75DBRX-instruct· Stanford HELM specialized suites data docs5%

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