Skip to content

Benchmark

HELM AIR-Bench: AIRBench 2024 - #35: Sowing Division

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
1.66

Score distribution

75.7%
Mean
86.7%
Median
26.9%
Std dev
30%
10th percentile
100%
90th percentile
3.3%
Min
3.3%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.66 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 Haiku· Stanford HELM specialized suites data docs100%
2Claude 3 Opus· Stanford HELM specialized suites data docs100%
3Claude 3 Sonnet· Stanford HELM specialized suites data docs100%
4Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs100%
5Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs100%
6Claude Haiku 4.5· Stanford HELM specialized suites data docs100%
7Claude Opus 4· Stanford HELM specialized suites data docs100%
8Claude Sonnet 4.5· Stanford HELM specialized suites data docs100%
9GPT 5· Stanford HELM specialized suites data docs100%
10GPT OSS 120B· Stanford HELM specialized suites data docs100%
11GPT OSS 20b· Stanford HELM specialized suites data docs100%
12GPT-5 nano· Stanford HELM specialized suites data docs100%
13o1· Stanford HELM specialized suites data docs100%
14o3 Mini· Stanford HELM specialized suites data docs100%
15OpenAI o3· Stanford HELM specialized suites data docs100%
16Qwen3 235B A22B Instruct 2507· Stanford HELM specialized suites data docs100%
17Qwen3 Next 80B A3Bthinking· Stanford HELM specialized suites data docs100%
18Claude Sonnet 4· Stanford HELM specialized suites data docs96.7%
19Gemini 2.5 Flash-Lite· Stanford HELM specialized suites data docs96.7%
20GPT 5.1· Stanford HELM specialized suites data docs96.7%
21GPT-4.5 Preview· Stanford HELM specialized suites data docs96.7%
22GPT-5 mini· Stanford HELM specialized suites data docs96.7%
23Granite 3.3 8b Instruct· Stanford HELM specialized suites data docs96.7%
24Kimi K2 Instruct· Stanford HELM specialized suites data docs96.7%
25Palmyra X5· Stanford HELM specialized suites data docs96.7%
26Claude Sonnet 3.7· Stanford HELM specialized suites data docs93.3%
27Gemini 2.0 Flash-Lite Preview· Stanford HELM specialized suites data docs93.3%
28Gemini 2.5 Pro Preview 03-25· Stanford HELM specialized suites data docs93.3%
29GPT 4.1· Stanford HELM specialized suites data docs93.3%
30GPT-4 Turbo· Stanford HELM specialized suites data docs93.3%
31Llama 4 Maverick· Stanford HELM specialized suites data docs93.3%
32o4 Mini· Stanford HELM specialized suites data docs93.3%
33Gemini 3 Pro Preview· Stanford HELM specialized suites data docs90%
34gemini-2.0-pro-exp-02-05· Stanford HELM specialized suites data docs90%
35GPT-4.1 nano· Stanford HELM specialized suites data docs90%
36Llama3 8B Chat· Stanford HELM specialized suites data docs90%
37Granite 4.0 H Small· Stanford HELM specialized suites data docs86.7%
38Llama 3.1 8B Instruct Turbo· Stanford HELM specialized suites data docs86.7%
39Gemini 2.0 Flash· Stanford HELM specialized suites data docs83.3%
40Qwen2 72B Instruct· Stanford HELM specialized suites data docs83.3%
41Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs83.3%
42gemini-2.5-flash-preview-04-17· Stanford HELM specialized suites data docs80%
43GLM 4.5 Air· Stanford HELM specialized suites data docs80%
44Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs80%
45GPT 4· Stanford HELM specialized suites data docs78.3%
46gemini-1.5-flash-002· Stanford HELM specialized suites data docs76.7%
47GPT-4.1 mini· Stanford HELM specialized suites data docs76.7%
48GPT-4o mini· Stanford HELM specialized suites data docs76.7%
49Yi 34B Chat· Stanford HELM specialized suites data docs76.7%
50gemini-1.5-pro-002· Stanford HELM specialized suites data docs73.3%

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