Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #36: Misrepresentation

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
1.19

Score distribution

56%
Mean
48.9%
Median
25.6%
Std dev
26.7%
10th percentile
91.1%
90th percentile
10%
Min
10%98.9%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.19 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-flash-002· Stanford HELM specialized suites data docs42.2%
52Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs41.1%
53Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs40%
54GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs38.9%
55GPT-4o mini· Stanford HELM specialized suites data docs38.9%
56Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs37.8%
57DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs34.4%
58GPT 4osetting not named by the source· Stanford HELM specialized suites data docs34.4%
59Yi 34B Chat· Stanford HELM specialized suites data docs34.4%
60GPT 4osetting not named by the source· Stanford HELM specialized suites data docs33.3%
61Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs33.3%
62Llama 4 Scout· Stanford HELM specialized suites data docs32.2%
63Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs31.1%
64Grok 4· Stanford HELM specialized suites data docs30%
65Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs28.9%
66O1 Mini· Stanford HELM specialized suites data docs27.8%
67Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs27.8%
68GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs26.7%
69Command R+· Stanford HELM specialized suites data docs22.2%
70DeepSeek V3· Stanford HELM specialized suites data docs22.2%
71Mistral Small· Stanford HELM specialized suites data docs22.2%
72Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs16.7%
73Mistral Large2407· Stanford HELM specialized suites data docs14.4%
74DBRX-instruct· Stanford HELM specialized suites data docs12.2%
75Command R Plus· Stanford HELM specialized suites data docs10%

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