Skip to content

Benchmark

HELM MMLU: Sociology - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
2.20

Score distribution

87.1%
Mean
90%
Median
9%
Std dev
80.6%
10th percentile
94%
90th percentile
38.3%
Min
38.3%96%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.20 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
1Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs96%
2Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs95.5%
3DeepSeek V3· Stanford HELM specialized suites data docs95.5%
4gemini-1.5-pro-002· Stanford HELM specialized suites data docs95%
5GPT 4osetting not named by the source· Stanford HELM specialized suites data docs94.5%
6Claude 3 Opus· Stanford HELM specialized suites data docs94%
7GPT 4osetting not named by the source· Stanford HELM specialized suites data docs94%
8Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs94%
9Qwen2 72B Instruct· Stanford HELM specialized suites data docs93.5%
10GPT 4· Stanford HELM specialized suites data docs93%
11Jamba 1.5 Large· Stanford HELM specialized suites data docs93%
12Meta Llama 3 70b· Stanford HELM specialized suites data docs93%
13Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs93%
14GPT 4 1106 Preview· Stanford HELM specialized suites data docs92.5%
15Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs92.5%
16gemini-1.5-pro-001· Stanford HELM specialized suites data docs92%
17Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs92%
18Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs92%
19Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs92%
20mixtral-8x22b· Stanford HELM specialized suites data docs92%
21gemini-1.5-flash-001· Stanford HELM specialized suites data docs91.5%
22GPT-4 Turbo· Stanford HELM specialized suites data docs91.5%
23Mistral Large2407· Stanford HELM specialized suites data docs91%
24text-unicorn· Stanford HELM specialized suites data docs91%
25Claude 3 Sonnet· Stanford HELM specialized suites data docs90.5%
26Claude Haiku 3.5· Stanford HELM specialized suites data docs90.5%
27Nova Prov1· Stanford HELM specialized suites data docs90.5%
28Claude 3 Haiku· Stanford HELM specialized suites data docs90%
29gemma-2-9b· Stanford HELM specialized suites data docs90%
30gemma2-27b· Stanford HELM specialized suites data docs90%
31GPT-4o mini· Stanford HELM specialized suites data docs90%
32Llama 2 70b· Stanford HELM specialized suites data docs90%
33Qwen1.5-72B· Stanford HELM specialized suites data docs90%
34Yi-34B· Stanford HELM specialized suites data docs90%
35DBRX-instruct· Stanford HELM specialized suites data docs89.6%
36Nova Litev1· Stanford HELM specialized suites data docs89.6%
37Jamba Instruct· Stanford HELM specialized suites data docs89.1%
38Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs89.1%
39Snowflake Arctic Instruct· Stanford HELM specialized suites data docs89.1%
40Claude 2.1· Stanford HELM specialized suites data docs88.6%
41Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs88.6%
42Command R Plus· Stanford HELM specialized suites data docs87.6%
43DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs87.6%
44Jamba 1.5 Mini· Stanford HELM specialized suites data docs87.6%
45GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs87.1%
46Mistral Small· Stanford HELM specialized suites data docs87.1%
47Command R+· Stanford HELM specialized suites data docs86.6%
48Meta Llama 3 8b· Stanford HELM specialized suites data docs86.6%
49Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs86.6%
50GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs86.1%

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