Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #13: Harassment

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
0.87

Score distribution

63.4%
Mean
64%
Median
21.7%
Std dev
32.3%
10th percentile
91.4%
90th percentile
13.4%
Min
13.4%98.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.87 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
51gemini-1.5-pro-002· Stanford HELM specialized suites data docs54.6%
52gemini-1.5-flash-002· Stanford HELM specialized suites data docs53.8%
53DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs52.4%
54Yi 34B Chat· Stanford HELM specialized suites data docs52.4%
55GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs52.2%
56Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs50.3%
57GPT-4o mini· Stanford HELM specialized suites data docs48.9%
58Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs44.9%
59DeepSeek R1· Stanford HELM specialized suites data docs43.5%
60GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs41.4%
61Grok 3 Mini· Stanford HELM specialized suites data docs39.8%
62Llama 4 Scout· Stanford HELM specialized suites data docs39%
63Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs36%
64O1 Mini· Stanford HELM specialized suites data docs36%
65DeepSeek V3· Stanford HELM specialized suites data docs35.8%
66Grok 3· Stanford HELM specialized suites data docs35.5%
67Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs34.9%
68Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs32.3%
69Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs28.2%
70Grok 4· Stanford HELM specialized suites data docs27.7%
71Command R+· Stanford HELM specialized suites data docs26.6%
72Mistral Large2407· Stanford HELM specialized suites data docs26.3%
73Mistral Small· Stanford HELM specialized suites data docs24.5%
74Command R Plus· Stanford HELM specialized suites data docs20.4%
75DBRX-instruct· Stanford HELM specialized suites data docs13.4%

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