Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #4: Automated Decision-Making

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
0.72

Score distribution

34.9%
Mean
30%
Median
19.2%
Std dev
12.7%
10th percentile
66.3%
90th percentile
9%
Min
9%81%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.72 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
51Llama3 8B Chat· Stanford HELM specialized suites data docs22%
52GLM 4.5 Air· Stanford HELM specialized suites data docs21.3%
53GPT-4.1 mini· Stanford HELM specialized suites data docs21.3%
54DBRX-instruct· Stanford HELM specialized suites data docs20.3%
55Qwen2 72B Instruct· Stanford HELM specialized suites data docs19.7%
56Llama 4 Scout· Stanford HELM specialized suites data docs19.3%
57GPT 4.1· Stanford HELM specialized suites data docs18.3%
58Kimi K2 Instruct· Stanford HELM specialized suites data docs18%
59Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs17.7%
60o3 Mini· Stanford HELM specialized suites data docs17.3%
61GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs16.7%
62Grok 3· Stanford HELM specialized suites data docs16.7%
63Grok 3 Mini· Stanford HELM specialized suites data docs16.3%
64Mistral Large2407· Stanford HELM specialized suites data docs16%
65Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs15.7%
66Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs15%
67Command R Plus· Stanford HELM specialized suites data docs14%
68Command R+· Stanford HELM specialized suites data docs12.7%
69DeepSeek R1· Stanford HELM specialized suites data docs12.3%
70GPT 4osetting not named by the source· Stanford HELM specialized suites data docs12.3%
71Grok 4· Stanford HELM specialized suites data docs12.3%
72Mistral Small· Stanford HELM specialized suites data docs12%
73GPT-4.1 nano· Stanford HELM specialized suites data docs11.7%
74Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs11.3%
75DeepSeek V3· Stanford HELM specialized suites data docs9%

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