Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #7: Supporting Malicious Organized Groups

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

66.3%
Mean
68.9%
Median
29.6%
Std dev
17.8%
10th percentile
97.8%
90th percentile
2.2%
Min
2.2%98.9%
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
51Llama 4 Scout· Stanford HELM specialized suites data docs57.8%
52gemini-1.5-flash-002· Stanford HELM specialized suites data docs55.6%
53Yi 34B Chat· Stanford HELM specialized suites data docs50%
54DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs48.9%
55GPT 4osetting not named by the source· Stanford HELM specialized suites data docs48.9%
56gemini-1.5-pro-002· Stanford HELM specialized suites data docs44.4%
57Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs43.3%
58Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs42.2%
59Grok 3 Mini· Stanford HELM specialized suites data docs38.9%
60GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs37.8%
61O1 Mini· Stanford HELM specialized suites data docs37.8%
62DeepSeek R1· Stanford HELM specialized suites data docs35.6%
63GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs32.2%
64DeepSeek V3· Stanford HELM specialized suites data docs28.9%
65Grok 3· Stanford HELM specialized suites data docs27.8%
66Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs27.8%
67Command R+· Stanford HELM specialized suites data docs17.8%
68Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs17.8%
69Mistral Small· Stanford HELM specialized suites data docs16.7%
70Grok 4· Stanford HELM specialized suites data docs13.3%
71Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs7.8%
72Mistral Large2407· Stanford HELM specialized suites data docs6.7%
73Command R Plus· Stanford HELM specialized suites data docs5.6%
74DBRX-instruct· Stanford HELM specialized suites data docs2.2%
75Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs2.2%

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