Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Defamation

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
1.04

Score distribution

68.1%
Mean
64.8%
Median
21%
Std dev
38.9%
10th percentile
97.2%
90th percentile
22.2%
Min
22.2%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.04 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 docs57.4%
52GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs57.4%
53Grok 4· Stanford HELM specialized suites data docs55.6%
54gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs54.6%
55GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs53.7%
56Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs53.7%
57Yi 34B Chat· Stanford HELM specialized suites data docs53.7%
58GPT 4· Stanford HELM specialized suites data docs52.8%
59Grok 3· Stanford HELM specialized suites data docs51.9%
60Llama 4 Scout· Stanford HELM specialized suites data docs51.9%
61GPT-4o mini· Stanford HELM specialized suites data docs50.9%
62Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs50.9%
63Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs45.4%
64gemini-1.5-flash-002· Stanford HELM specialized suites data docs44.4%
65GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs44.4%
66O1 Mini· Stanford HELM specialized suites data docs43.5%
67DeepSeek V3· Stanford HELM specialized suites data docs38.9%
68Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs38.9%
69Mistral Large2407· Stanford HELM specialized suites data docs37%
70Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs37%
71Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs34.3%
72Mistral Small· Stanford HELM specialized suites data docs34.3%
73Command R+· Stanford HELM specialized suites data docs33.3%
74DBRX-instruct· Stanford HELM specialized suites data docs24.1%
75Command R Plus· Stanford HELM specialized suites data docs22.2%

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