Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Deception

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
1.01

Score distribution

60.8%
Mean
60.7%
Median
25.7%
Std dev
25.2%
10th percentile
93%
90th percentile
7.4%
Min
7.4%97.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.01 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
51GPT 4osetting not named by the source· Stanford HELM specialized suites data docs49.6%
52GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs47.8%
53Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs47.4%
54DeepSeek R1· Stanford HELM specialized suites data docs46.3%
55GPT-4o mini· Stanford HELM specialized suites data docs45.9%
56GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs44.4%
57Grok 3 Mini· Stanford HELM specialized suites data docs39.3%
58DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs35.2%
59O1 Mini· Stanford HELM specialized suites data docs34.8%
60GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs34.1%
61Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs32.2%
62Yi 34B Chat· Stanford HELM specialized suites data docs31.9%
63Grok 3· Stanford HELM specialized suites data docs31.5%
64Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs28.9%
65Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs27.8%
66DeepSeek V3· Stanford HELM specialized suites data docs25.9%
67Grok 4· Stanford HELM specialized suites data docs25.9%
68GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs25.2%
69Mistral Large2407· Stanford HELM specialized suites data docs21.1%
70Mistral Small· Stanford HELM specialized suites data docs20.7%
71Command R+· Stanford HELM specialized suites data docs17.8%
72Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs17.8%
73Command R Plus· Stanford HELM specialized suites data docs15.6%
74Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs11.5%
75DBRX-instruct· Stanford HELM specialized suites data docs7.4%

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