Skip to content

Benchmark

HELM Safety: BBQ - BBQ (unambiguous)

Safety benchmarks examine risks in model behaviour, including how models respond to unsafe requests. Scores are percentages. Lower is better.

Models measured
69
Scores recorded
74
Sources
1
IQ weight
1.17

Score distribution

-19.6%
Mean
-19.4%
Median
2.6%
Std dev
-21.4%
10th percentile
-16.9%
90th percentile
-34.6%
Min
-34.6%-11.9%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.17 for chat models, measured over 69 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 2.0 Flash· Stanford HELM specialized suites data docs-19%
52DeepSeek V3· Stanford HELM specialized suites data docs-18.8%
53gemini-1.5-flash-001· Stanford HELM specialized suites data docs-18.6%
54Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs-18.5%
55Llama3 8B Chat· Stanford HELM specialized suites data docs-18.5%
56Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs-18.5%
57Command R Plus· Stanford HELM specialized suites data docs-18.3%
58Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs-18.2%
59Mistral Small· Stanford HELM specialized suites data docs-18.1%
60DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs-18.1%
61GPT-4.1 mini· Stanford HELM specialized suites data docs-18%
62GPT-5 nano· Stanford HELM specialized suites data docs-18%
63Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs-17.9%
64GPT-4.1 nano· Stanford HELM specialized suites data docs-17.9%
65GPT-4o mini· Stanford HELM specialized suites data docs-17.6%
66Claude 3 Haiku· Stanford HELM specialized suites data docs-17.2%
67GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs-16.9%
68Llama 4 Scout· Stanford HELM specialized suites data docs-16.8%
69GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs-16.8%
70Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs-16.7%
71Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs-16.4%
72Granite 4.0 H Small· Stanford HELM specialized suites data docs-15.7%
73GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs-15.6%
74Mistral (7B) Instruct v0.1· Stanford HELM specialized suites data docs-11.9%

Showing 51–74 of 74 results.

Data credits: Stanford HELM (Stanford CRFM); scenario scores from Capabilities, Safety, Long Context, MedHELM, AIR-Bench, MMLU and Finance