Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - Child Harm

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
1.08

Score distribution

57.8%
Mean
58.1%
Median
22.3%
Std dev
23.8%
10th percentile
85.7%
90th percentile
6.7%
Min
6.7%92.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.08 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
51GLM 4.5 Air· Stanford HELM specialized suites data docs48.6%
52Grok 3 Mini· Stanford HELM specialized suites data docs47.1%
53GPT 4osetting not named by the source· Stanford HELM specialized suites data docs46.2%
54GPT 4osetting not named by the source· Stanford HELM specialized suites data docs45.2%
55Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs45.2%
56GPT-4o mini· Stanford HELM specialized suites data docs44.3%
57DeepSeek R1· Stanford HELM specialized suites data docs41%
58O1 Mini· Stanford HELM specialized suites data docs40.5%
59Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs38.6%
60Yi 34B Chat· Stanford HELM specialized suites data docs36.7%
61GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs36.2%
62Grok 3· Stanford HELM specialized suites data docs34.8%
63DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs32.9%
64Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs31%
65Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs31%
66DeepSeek V3· Stanford HELM specialized suites data docs29.5%
67Mistral Large2407· Stanford HELM specialized suites data docs26.7%
68GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs23.8%
69Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs22.9%
70Command R+· Stanford HELM specialized suites data docs19.5%
71Command R Plus· Stanford HELM specialized suites data docs18.6%
72Mistral Small· Stanford HELM specialized suites data docs17.1%
73Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs17.1%
74DBRX-instruct· Stanford HELM specialized suites data docs8.6%
75Mistral (7B) Instruct v0.3· 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