Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Refusal Rate

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
1.07

Score distribution

65%
Mean
66.2%
Median
17.4%
Std dev
40.8%
10th percentile
86.7%
90th percentile
25.4%
Min
25.4%93.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.07 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
51GLM 4.5 Air· Stanford HELM specialized suites data docs57.1%
52GPT-4o mini· Stanford HELM specialized suites data docs56.3%
53Yi 34B Chat· Stanford HELM specialized suites data docs53.6%
54Grok 3 Mini· Stanford HELM specialized suites data docs53.5%
55DeepSeek R1· Stanford HELM specialized suites data docs52.9%
56GPT 4osetting not named by the source· Stanford HELM specialized suites data docs52.8%
57GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs52.5%
58Llama 4 Scout· Stanford HELM specialized suites data docs52.3%
59Grok 3· Stanford HELM specialized suites data docs51.3%
60DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs50.5%
61Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs48.6%
62Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs47%
63O1 Mini· Stanford HELM specialized suites data docs45.2%
64Grok 4· Stanford HELM specialized suites data docs44.4%
65Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs44%
66GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs44%
67Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs42.5%
68DeepSeek V3· Stanford HELM specialized suites data docs40.8%
69Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs39.1%
70Mistral Large2407· Stanford HELM specialized suites data docs35.3%
71Mistral Small· Stanford HELM specialized suites data docs32.8%
72Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs32.6%
73Command R+· Stanford HELM specialized suites data docs31.8%
74Command R Plus· Stanford HELM specialized suites data docs29.3%
75DBRX-instruct· Stanford HELM specialized suites data docs25.4%

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