Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #34: Mis/disinformation

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
1.12

Score distribution

60.9%
Mean
66.7%
Median
26.4%
Std dev
20%
10th percentile
90%
90th percentile
6.7%
Min
6.7%96.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.12 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 docs53.3%
52DeepSeek R1· Stanford HELM specialized suites data docs51.7%
53GPT 4· Stanford HELM specialized suites data docs45%
54DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs43.3%
55GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs41.7%
56GPT-4o mini· Stanford HELM specialized suites data docs40%
57Grok 3· Stanford HELM specialized suites data docs36.7%
58Grok 3 Mini· Stanford HELM specialized suites data docs36.7%
59GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs33.3%
60Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs30%
61Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs26.7%
62O1 Mini· Stanford HELM specialized suites data docs26.7%
63Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs26.7%
64DeepSeek V3· Stanford HELM specialized suites data docs23.3%
65GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs23.3%
66Yi 34B Chat· Stanford HELM specialized suites data docs23.3%
67Grok 4· Stanford HELM specialized suites data docs20%
68Mistral Small· Stanford HELM specialized suites data docs20%
69Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs20%
70Command R Plus· Stanford HELM specialized suites data docs16.7%
71GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs16.7%
72Mistral Large2407· Stanford HELM specialized suites data docs13.3%
73Command R+· Stanford HELM specialized suites data docs10%
74Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs10%
75DBRX-instruct· Stanford HELM specialized suites data docs6.7%

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