Skip to content

Benchmark

HELM MMLU: Moral Scenarios - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
1.23

Score distribution

57.7%
Mean
57.9%
Median
19.7%
Std dev
33%
10th percentile
83.4%
90th percentile
23.1%
Min
23.1%90.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.23 for chat models, measured over 64 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1GPT 4· Stanford HELM specialized suites data docs90.2%
2Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs88.8%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs88.2%
4Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs87.6%
5GPT 4osetting not named by the source· Stanford HELM specialized suites data docs84.1%
6Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs84.1%
7Mistral Large2407· Stanford HELM specialized suites data docs83.9%
8Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs83.4%
9Claude 3 Opus· Stanford HELM specialized suites data docs82.6%
10GPT 4 1106 Preview· Stanford HELM specialized suites data docs81.6%
11Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs81.5%
12Qwen2 72B Instruct· Stanford HELM specialized suites data docs81.5%
13DeepSeek V3· Stanford HELM specialized suites data docs80.8%
14GPT-4 Turbo· Stanford HELM specialized suites data docs80.3%
15GPT 4osetting not named by the source· Stanford HELM specialized suites data docs80.2%
16gemini-1.5-pro-002· Stanford HELM specialized suites data docs79.2%
17Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs78.7%
18Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs78.3%
19Nova Prov1· Stanford HELM specialized suites data docs76%
20gemini-1.5-pro-001· Stanford HELM specialized suites data docs73.9%
21Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs71.1%
22Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs69.8%
23Nova Litev1· Stanford HELM specialized suites data docs69.4%
24Jamba 1.5 Large· Stanford HELM specialized suites data docs68.6%
25gemini-1.5-flash-002· Stanford HELM specialized suites data docs67.6%
26Qwen1.5-72B· Stanford HELM specialized suites data docs66.9%
27mixtral-8x22b· Stanford HELM specialized suites data docs64.6%
28Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs63.9%
29gemini-1.5-flash-001· Stanford HELM specialized suites data docs63.7%
30Claude 3 Sonnet· Stanford HELM specialized suites data docs62.6%
31Yi-34B· Stanford HELM specialized suites data docs60.6%
32Meta Llama 3 70b· Stanford HELM specialized suites data docs59.8%
33Command R Plus· Stanford HELM specialized suites data docs58.5%
34Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs57.9%
35Mistral Small· Stanford HELM specialized suites data docs57.5%
36text-unicorn· Stanford HELM specialized suites data docs56.2%
37DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs54.4%
38Claude 2.1· Stanford HELM specialized suites data docs52%
39Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs51.1%
40Claude 3 Haiku· Stanford HELM specialized suites data docs50.2%
41Claude Instant 1.2· Stanford HELM specialized suites data docs48.8%
42GPT-4o mini· Stanford HELM specialized suites data docs48.5%
43Claude Haiku 3.5· Stanford HELM specialized suites data docs47.6%
44DBRX-instruct· Stanford HELM specialized suites data docs46.5%
45Jamba Instruct· Stanford HELM specialized suites data docs46.5%
46Nova Microv1· Stanford HELM specialized suites data docs46.4%
47Command R+· Stanford HELM specialized suites data docs45.1%
48Llama 2 70b· Stanford HELM specialized suites data docs45%
49Meta Llama 3 8b· Stanford HELM specialized suites data docs41.6%
50Llama 2 13b· Stanford HELM specialized suites data docs40.7%

Showing 1–50 of 68 results.

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