Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #20: Monetized

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
0.61

Score distribution

38.6%
Mean
26.7%
Median
25.9%
Std dev
13.3%
10th percentile
81.7%
90th percentile
10%
Min
10%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.61 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
1Palmyra X5· Stanford HELM specialized suites data docs100%
2Claude 3 Haiku· Stanford HELM specialized suites data docs96.7%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs90%
4Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs90%
5Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs88.3%
6Claude 3 Sonnet· Stanford HELM specialized suites data docs86.7%
7Claude Haiku 4.5· Stanford HELM specialized suites data docs86.7%
8Claude 3 Opus· Stanford HELM specialized suites data docs81.7%
9Llama 4 Maverick· Stanford HELM specialized suites data docs76.7%
10Llama3 8B Chat· Stanford HELM specialized suites data docs75%
11GPT OSS 20b· Stanford HELM specialized suites data docs73.3%
12gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs68.3%
13Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs66.7%
14GPT 5.1· Stanford HELM specialized suites data docs66.7%
15Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs66.7%
16Granite 4.0 H Small· Stanford HELM specialized suites data docs66.7%
17Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs63.3%
18GPT-4.5 Preview· Stanford HELM specialized suites data docs61.7%
19GPT OSS 120B· Stanford HELM specialized suites data docs60%
20Claude Sonnet 4· Stanford HELM specialized suites data docs56.7%
21Claude Sonnet 4.5· Stanford HELM specialized suites data docs56.7%
22Gemini 2.0 Flash· Stanford HELM specialized suites data docs55%
23Claude Opus 4· Stanford HELM specialized suites data docs53.3%
24gemini-1.5-pro-002· Stanford HELM specialized suites data docs53.3%
25gemini-1.5-flash-002· Stanford HELM specialized suites data docs48.3%
26Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs48.3%
27Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs45%
28Claude Sonnet 3.7· Stanford HELM specialized suites data docs43.3%
29Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs41.7%
30Kimi K2 Instruct· Stanford HELM specialized suites data docs38.3%
31Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs38.3%
32GPT-4 Turbo· Stanford HELM specialized suites data docs33.3%
33gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs28.3%
34GPT 5· Stanford HELM specialized suites data docs28.3%
35GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs28.3%
36GPT-4.1 nano· Stanford HELM specialized suites data docs28.3%
37Qwen2 72B Instruct· Stanford HELM specialized suites data docs28.3%
38o4 Mini· Stanford HELM specialized suites data docs26.7%
39OpenAI o3· Stanford HELM specialized suites data docs26.7%
40Gemini 3 Pro Preview· Stanford HELM specialized suites data docs25%
41GLM 4.5 Air· Stanford HELM specialized suites data docs25%
42GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs25%
43GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs23.3%
44Llama 4 Scout· Stanford HELM specialized suites data docs23.3%
45Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs23.3%
46Yi 34B Chat· Stanford HELM specialized suites data docs23.3%
47DeepSeek R1· Stanford HELM specialized suites data docs21.7%
48GPT-5 mini· Stanford HELM specialized suites data docs21.7%
49Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs21.7%
50GPT 4osetting not named by the source· Stanford HELM specialized suites data docs20%

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