Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Discrimination/Bias

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
1.18

Score distribution

64%
Mean
61.8%
Median
14%
Std dev
45.8%
10th percentile
85.5%
90th percentile
41.9%
Min
41.9%90.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.18 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 4· Stanford HELM specialized suites data docs55.9%
52Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs54.7%
53Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs54.4%
54Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs54.3%
55GPT-4.1 mini· Stanford HELM specialized suites data docs54%
56Yi 34B Chat· Stanford HELM specialized suites data docs53.9%
57Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs52.1%
58GPT 4osetting not named by the source· Stanford HELM specialized suites data docs51.9%
59DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs51.7%
60GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs50.5%
61GPT-4.1 nano· Stanford HELM specialized suites data docs49.8%
62Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs49%
63Llama 4 Scout· Stanford HELM specialized suites data docs48.8%
64Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs48.8%
65Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs48.4%
66GPT-4o mini· Stanford HELM specialized suites data docs47.9%
67GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs46.4%
68DeepSeek V3· Stanford HELM specialized suites data docs45.8%
69Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs45.2%
70Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs44.9%
71Mistral Small· Stanford HELM specialized suites data docs43.2%
72Mistral Large2407· Stanford HELM specialized suites data docs42.8%
73Command R+· Stanford HELM specialized suites data docs42.3%
74DBRX-instruct· Stanford HELM specialized suites data docs42%
75Command R Plus· Stanford HELM specialized suites data docs41.9%

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