Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #27: Disrupting Social Order (China-unique)

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
1.46

Score distribution

71.6%
Mean
81.4%
Median
27.2%
Std dev
25.6%
10th percentile
98.1%
90th percentile
3.1%
Min
3.1%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.46 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 docs65.6%
52GPT-4o mini· Stanford HELM specialized suites data docs64.7%
53DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs62.8%
54GPT 4osetting not named by the source· Stanford HELM specialized suites data docs61.7%
55Llama 4 Scout· Stanford HELM specialized suites data docs58.9%
56Yi 34B Chat· Stanford HELM specialized suites data docs58.9%
57Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs54.2%
58Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs52.5%
59GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs51.7%
60GPT 4osetting not named by the source· Stanford HELM specialized suites data docs49.7%
61Grok 3 Mini· Stanford HELM specialized suites data docs45%
62DeepSeek V3· Stanford HELM specialized suites data docs43.6%
63Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs38.6%
64GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs35.6%
65Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs31.9%
66Grok 3· Stanford HELM specialized suites data docs28.9%
67O1 Mini· Stanford HELM specialized suites data docs26.1%
68Grok 4· Stanford HELM specialized suites data docs25.6%
69Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs25%
70Command R+· Stanford HELM specialized suites data docs19.7%
71Mistral Large2407· Stanford HELM specialized suites data docs18.3%
72Command R Plus· Stanford HELM specialized suites data docs11.4%
73Mistral Small· Stanford HELM specialized suites data docs11.1%
74Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs8.6%
75DBRX-instruct· Stanford HELM specialized suites data docs3.1%

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