Skip to content

Benchmark

HELM Safety: HarmBench - LM Evaluated Safety score

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

Models measured
69
Scores recorded
74
Sources
1
IQ weight
0.89

Score distribution

74.4%
Mean
76.8%
Median
19.7%
Std dev
46.9%
10th percentile
97.6%
90th percentile
23.3%
Min
23.3%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.89 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-pro-exp-02-05· Stanford HELM specialized suites data docs65.3%
52DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs64.9%
53Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs64.8%
54GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs63.3%
55Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs62.7%
56gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs62.6%
57Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs61.6%
58Llama 4 Scout· Stanford HELM specialized suites data docs60%
59Grok 3 Mini· Stanford HELM specialized suites data docs57.2%
60GLM 4.5 Air· Stanford HELM specialized suites data docs56.1%
61DeepSeek R1setting not named by the source· Stanford HELM specialized suites data docs54.6%
62Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs52.1%
63Command R+· Stanford HELM specialized suites data docs50.4%
64DeepSeek V3· Stanford HELM specialized suites data docs49.7%
65Command R Plus· Stanford HELM specialized suites data docs48.5%
66DeepSeek R1setting not named by the source· Stanford HELM specialized suites data docs47.1%
67Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs46.9%
68Grok 3· Stanford HELM specialized suites data docs45.3%
69Mistral Small· Stanford HELM specialized suites data docs45.2%
70Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs45.1%
71Grok 4· Stanford HELM specialized suites data docs39.7%
72Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs34.4%
73DBRX-instruct· Stanford HELM specialized suites data docs27.1%
74Mistral (7B) Instruct v0.1· Stanford HELM specialized suites data docs23.3%

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