Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #15: Perpetuating Harmful Beliefs

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
0.91

Score distribution

61.9%
Mean
61.1%
Median
16.6%
Std dev
41.3%
10th percentile
84.1%
90th percentile
22.2%
Min
22.2%95.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.91 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
51Yi 34B Chat· Stanford HELM specialized suites data docs53.2%
52GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs52.4%
53GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs50.8%
54GPT-4.1 nano· Stanford HELM specialized suites data docs50.8%
55Llama 4 Scout· Stanford HELM specialized suites data docs48.4%
56Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs48.4%
57GPT 4osetting not named by the source· Stanford HELM specialized suites data docs47.6%
58Grok 3 Mini· Stanford HELM specialized suites data docs47.6%
59Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs46.8%
60DeepSeek V3· Stanford HELM specialized suites data docs46%
61GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs46%
62GPT 4· Stanford HELM specialized suites data docs45.2%
63O1 Mini· Stanford HELM specialized suites data docs45.2%
64GPT-4o mini· Stanford HELM specialized suites data docs44.4%
65Grok 3· Stanford HELM specialized suites data docs42.9%
66Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs42.9%
67Grok 4· Stanford HELM specialized suites data docs42.1%
68Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs41.3%
69Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs39.7%
70Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs39.7%
71DBRX-instruct· Stanford HELM specialized suites data docs37.3%
72Command R+· Stanford HELM specialized suites data docs33.3%
73Mistral Large2407· Stanford HELM specialized suites data docs31%
74Mistral Small· Stanford HELM specialized suites data docs31%
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