Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Manipulation

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

63.9%
Mean
63.3%
Median
25%
Std dev
27.3%
10th percentile
94.7%
90th percentile
8.7%
Min
8.7%99.3%
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
51GPT-4o mini· Stanford HELM specialized suites data docs54%
52DeepSeek R1· Stanford HELM specialized suites data docs52.7%
53Yi 34B Chat· Stanford HELM specialized suites data docs51.3%
54GPT 4osetting not named by the source· Stanford HELM specialized suites data docs48%
55GPT 4osetting not named by the source· Stanford HELM specialized suites data docs47.3%
56Grok 3 Mini· Stanford HELM specialized suites data docs47.3%
57Llama 4 Scout· Stanford HELM specialized suites data docs46%
58DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs44.7%
59GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs44.7%
60Grok 3· Stanford HELM specialized suites data docs43.3%
61Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs40.7%
62Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs40%
63Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs40%
64Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs36.7%
65GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs32%
66Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs32%
67Grok 4· Stanford HELM specialized suites data docs31.3%
68O1 Mini· Stanford HELM specialized suites data docs27.3%
69DeepSeek V3· Stanford HELM specialized suites data docs25.3%
70Command R+· Stanford HELM specialized suites data docs21.3%
71Mistral Small· Stanford HELM specialized suites data docs18.7%
72Mistral Large2407· Stanford HELM specialized suites data docs16.7%
73Command R Plus· Stanford HELM specialized suites data docs15.3%
74Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs15.3%
75DBRX-instruct· Stanford HELM specialized suites data docs8.7%

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