Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #12: Military and Warfare

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
2.08

Score distribution

64.5%
Mean
78.3%
Median
31.8%
Std dev
10%
10th percentile
98.3%
90th percentile
8.3%
Min
8.3%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.08 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-4.1 mini· Stanford HELM specialized suites data docs63.3%
52Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs56.7%
53Grok 3· Stanford HELM specialized suites data docs48.3%
54GPT 4osetting not named by the source· Stanford HELM specialized suites data docs46.7%
55GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs45%
56O1 Mini· Stanford HELM specialized suites data docs40%
57Yi 34B Chat· Stanford HELM specialized suites data docs31.7%
58Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs26.7%
59Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs20%
60DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs16.7%
61Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs16.7%
62Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs16.7%
63DeepSeek R1· Stanford HELM specialized suites data docs15%
64Grok 3 Mini· Stanford HELM specialized suites data docs15%
65Command R+· Stanford HELM specialized suites data docs13.3%
66Grok 4· Stanford HELM specialized suites data docs13.3%
67Command R Plus· Stanford HELM specialized suites data docs10%
68DBRX-instruct· Stanford HELM specialized suites data docs10%
69DeepSeek V3· Stanford HELM specialized suites data docs10%
70GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs10%
71GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs10%
72GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs10%
73Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs10%
74Mistral Large2407· Stanford HELM specialized suites data docs10%
75Mistral Small· Stanford HELM specialized suites data docs8.3%

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