Skip to content

Benchmark

HELM MMLU: Machine Learning - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
1.64

Score distribution

59.4%
Mean
61.6%
Median
13.3%
Std dev
41.1%
10th percentile
76.8%
90th percentile
28.6%
Min
28.6%83.9%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.64 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 docs83.9%
2gemini-1.5-pro-002· Stanford HELM specialized suites data docs83%
3Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs79.5%
4Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs78.6%
5DeepSeek V3· Stanford HELM specialized suites data docs78.6%
6GPT 4osetting not named by the source· Stanford HELM specialized suites data docs77.7%
7Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs77.7%
8GPT 4osetting not named by the source· Stanford HELM specialized suites data docs76.8%
9Qwen2 72B Instruct· Stanford HELM specialized suites data docs76.8%
10Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs75.9%
11GPT 4· Stanford HELM specialized suites data docs75.9%
12Claude 3 Opus· Stanford HELM specialized suites data docs74.1%
13GPT-4 Turbo· Stanford HELM specialized suites data docs74.1%
14GPT 4 1106 Preview· Stanford HELM specialized suites data docs72.3%
15Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs71.4%
16Meta Llama 3 70b· Stanford HELM specialized suites data docs71.4%
17Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs69.6%
18Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs69.6%
19Jamba 1.5 Large· Stanford HELM specialized suites data docs68.8%
20Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs68.8%
21Claude Instant 1.2· Stanford HELM specialized suites data docs67%
22Qwen1.5-72B· Stanford HELM specialized suites data docs67%
23Mistral Large2407· Stanford HELM specialized suites data docs66.1%
24mixtral-8x22b· Stanford HELM specialized suites data docs66.1%
25gemini-1.5-pro-001· Stanford HELM specialized suites data docs65.2%
26Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs65.2%
27Claude 3 Sonnet· Stanford HELM specialized suites data docs64.3%
28Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs63.4%
29DBRX-instruct· Stanford HELM specialized suites data docs62.5%
30gemma2-27b· Stanford HELM specialized suites data docs62.5%
31Nova Prov1· Stanford HELM specialized suites data docs62.5%
32text-unicorn· Stanford HELM specialized suites data docs62.5%
33gemini-1.5-flash-002· Stanford HELM specialized suites data docs61.6%
34GPT-4o mini· Stanford HELM specialized suites data docs61.6%
35Claude 3 Haiku· Stanford HELM specialized suites data docs58.9%
36Yi-34B· Stanford HELM specialized suites data docs58%
37gemini-1.5-flash-001· Stanford HELM specialized suites data docs57.1%
38DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs56.3%
39Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs56.3%
40Mistral Small· Stanford HELM specialized suites data docs56.3%
41Nova Microv1· Stanford HELM specialized suites data docs56.3%
42gemma-7b· Stanford HELM specialized suites data docs55.4%
43Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs55.4%
44Meta Llama 3 8b· Stanford HELM specialized suites data docs54.5%
45Jamba Instruct· Stanford HELM specialized suites data docs53.6%
46Claude Haiku 3.5· Stanford HELM specialized suites data docs51.8%
47Command R Plus· Stanford HELM specialized suites data docs51.8%
48gemma-2-9b· Stanford HELM specialized suites data docs50.9%
49Jamba 1.5 Mini· Stanford HELM specialized suites data docs50.9%
50Nova Litev1· Stanford HELM specialized suites data docs50.9%

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