Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #26: Deterring Democratic Participation

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
0.80

Score distribution

58.9%
Mean
56.7%
Median
25.2%
Std dev
23.3%
10th percentile
95%
90th percentile
10%
Min
10%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.80 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 docs45%
52Llama 4 Scout· Stanford HELM specialized suites data docs44.2%
53gemini-1.5-flash-002· Stanford HELM specialized suites data docs43.3%
54Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs43.3%
55DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs41.7%
56DeepSeek R1· Stanford HELM specialized suites data docs41.7%
57GPT 4osetting not named by the source· Stanford HELM specialized suites data docs41.7%
58GPT-4o mini· Stanford HELM specialized suites data docs40%
59Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs35%
60Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs35%
61Grok 3 Mini· Stanford HELM specialized suites data docs31.7%
62Grok 3· Stanford HELM specialized suites data docs30.8%
63GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs30%
64Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs30%
65Grok 4· Stanford HELM specialized suites data docs29.2%
66O1 Mini· Stanford HELM specialized suites data docs28.3%
67DeepSeek V3· Stanford HELM specialized suites data docs26.7%
68Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs23.3%
69Mistral Large2407· Stanford HELM specialized suites data docs22.5%
70Command R+· Stanford HELM specialized suites data docs20.8%
71GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs20%
72Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs20%
73Mistral Small· Stanford HELM specialized suites data docs18.3%
74Command R Plus· Stanford HELM specialized suites data docs11.7%
75DBRX-instruct· Stanford HELM specialized suites data docs10%

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