Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #10: Depicting Violence

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
0.50

Score distribution

43.9%
Mean
38%
Median
17.6%
Std dev
24.7%
10th percentile
70%
90th percentile
18%
Min
18%90.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.50 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
1Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs90.7%
2GPT OSS 120B· Stanford HELM specialized suites data docs86%
3GPT OSS 20b· Stanford HELM specialized suites data docs82%
4Claude Haiku 4.5· Stanford HELM specialized suites data docs75.3%
5Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs74%
6Claude Sonnet 4· Stanford HELM specialized suites data docs72.7%
7GPT-5 nano· Stanford HELM specialized suites data docs70.7%
8Claude 3 Haiku· Stanford HELM specialized suites data docs70%
9Palmyra X5· Stanford HELM specialized suites data docs70%
10Claude 3 Sonnet· Stanford HELM specialized suites data docs68%
11Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs68%
12GPT 5· Stanford HELM specialized suites data docs66.7%
13Llama3 8B Chat· Stanford HELM specialized suites data docs65.3%
14Claude Sonnet 4.5· Stanford HELM specialized suites data docs63.3%
15GPT-5 mini· Stanford HELM specialized suites data docs62%
16Claude 3 Opus· Stanford HELM specialized suites data docs60%
17Claude Opus 4· Stanford HELM specialized suites data docs59.3%
18Claude Sonnet 3.7· Stanford HELM specialized suites data docs58%
19Granite 4.0 H Small· Stanford HELM specialized suites data docs56.7%
20OpenAI o3· Stanford HELM specialized suites data docs56%
21GPT 5.1· Stanford HELM specialized suites data docs55.3%
22Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs54.7%
23Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs54%
24o4 Mini· Stanford HELM specialized suites data docs51.3%
25Llama 4 Maverick· Stanford HELM specialized suites data docs49.3%
26Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs49.3%
27gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs48.7%
28Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs45.3%
29GPT-4.5 Preview· Stanford HELM specialized suites data docs43.3%
30Gemini 2.0 Flash· Stanford HELM specialized suites data docs42.7%
31GPT-4 Turbo· Stanford HELM specialized suites data docs42.7%
32Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs42.7%
33GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs41.3%
34Kimi K2 Instruct· Stanford HELM specialized suites data docs40.7%
35DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs39.3%
36GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs38.7%
37GPT-4.1 nano· Stanford HELM specialized suites data docs38.7%
38GPT 4osetting not named by the source· Stanford HELM specialized suites data docs38%
39Qwen1.5-72B-Chat· Stanford HELM specialized suites data docs38%
40Mixtral 8x7B Instruct v0.1· Stanford HELM specialized suites data docs37.3%
41Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs36.7%
42Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs36.7%
43o1· Stanford HELM specialized suites data docs36%
44Yi 34B Chat· Stanford HELM specialized suites data docs36%
45Llama 4 Scout· Stanford HELM specialized suites data docs35.3%
46Mistral (7B) Instruct v0.3· Stanford HELM specialized suites data docs35.3%
47Mixtral 8x22B Instruct v0.1· Stanford HELM specialized suites data docs35.3%
48GPT 4· Stanford HELM specialized suites data docs34.7%
49Mistral Large2407· Stanford HELM specialized suites data docs34.7%
50Qwen2 72B Instruct· Stanford HELM specialized suites data docs34.7%

Showing 1–50 of 75 results.

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