Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Hate/Toxicity

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
0.96

Score distribution

75.5%
Mean
77.7%
Median
15.6%
Std dev
50.6%
10th percentile
94%
90th percentile
37.3%
Min
37.3%98.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.96 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
51DeepSeek R1· Stanford HELM specialized suites data docs70.9%
52Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs69.9%
53GPT 4osetting not named by the source· Stanford HELM specialized suites data docs68.2%
54Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs68.2%
55DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs65.4%
56Grok 3 Mini· Stanford HELM specialized suites data docs64.9%
57Yi 34B Chat· Stanford HELM specialized suites data docs64%
58Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs63.8%
59Llama 4 Scout· Stanford HELM specialized suites data docs62.4%
60GPT-4o mini· Stanford HELM specialized suites data docs61.6%
61Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs61%
62GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs59.4%
63DeepSeek V3· Stanford HELM specialized suites data docs58.6%
64Grok 3· Stanford HELM specialized suites data docs57.9%
65Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs54.9%
66Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs53%
67GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs50.7%
68Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs50.6%
69Mistral Large2407· Stanford HELM specialized suites data docs50.1%
70O1 Mini· Stanford HELM specialized suites data docs49.4%
71Grok 4· Stanford HELM specialized suites data docs48.3%
72Mistral Small· Stanford HELM specialized suites data docs47.6%
73Command R+· Stanford HELM specialized suites data docs44.9%
74DBRX-instruct· Stanford HELM specialized suites data docs40.7%
75Command R Plus· Stanford HELM specialized suites data docs37.3%

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