Skip to content

Benchmark

HELM MMLU: High School Statistics - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
1.70

Score distribution

65.1%
Mean
67.1%
Median
13.4%
Std dev
48.6%
10th percentile
79.6%
90th percentile
25.9%
Min
25.9%88.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.70 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
1gemini-1.5-pro-002· Stanford HELM specialized suites data docs88.4%
2DeepSeek V3· Stanford HELM specialized suites data docs86.1%
3gemini-1.5-pro-001· Stanford HELM specialized suites data docs85.6%
4Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs85.2%
5Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs84.7%
6Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs83.8%
7Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs80.1%
8Claude 3 Opus· Stanford HELM specialized suites data docs79.6%
9Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs79.6%
10GPT 4· Stanford HELM specialized suites data docs79.2%
11GPT 4osetting not named by the source· Stanford HELM specialized suites data docs78.7%
12Nova Prov1· Stanford HELM specialized suites data docs77.8%
13GPT 4osetting not named by the source· Stanford HELM specialized suites data docs76.9%
14Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs76.9%
15Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs76.4%
16GPT-4 Turbo· Stanford HELM specialized suites data docs75.9%
17Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs75.5%
18Qwen2 72B Instruct· Stanford HELM specialized suites data docs75%
19gemini-1.5-flash-001· Stanford HELM specialized suites data docs73.6%
20Meta Llama 3 70b· Stanford HELM specialized suites data docs73.6%
21text-unicorn· Stanford HELM specialized suites data docs73.1%
22gemini-1.5-flash-002· Stanford HELM specialized suites data docs72.7%
23Nova Litev1· Stanford HELM specialized suites data docs71.8%
24gemma2-27b· Stanford HELM specialized suites data docs71.3%
25Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs70.8%
26GPT-4o mini· Stanford HELM specialized suites data docs70.4%
27Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs69.9%
28Claude 2.1· Stanford HELM specialized suites data docs69.4%
29Claude 3 Sonnet· Stanford HELM specialized suites data docs69.4%
30Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs69.4%
31mixtral-8x22b· Stanford HELM specialized suites data docs68.5%
32gemma-2-9b· Stanford HELM specialized suites data docs67.6%
33Claude 3 Haiku· Stanford HELM specialized suites data docs67.1%
34DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs67.1%
35Nova Microv1· Stanford HELM specialized suites data docs67.1%
36Qwen1.5-72B· Stanford HELM specialized suites data docs67.1%
37Yi-34B· Stanford HELM specialized suites data docs67.1%
38GPT 4 1106 Preview· Stanford HELM specialized suites data docs66.2%
39DBRX-instruct· Stanford HELM specialized suites data docs65.7%
40Jamba 1.5 Large· Stanford HELM specialized suites data docs65.7%
41Mistral Large2407· Stanford HELM specialized suites data docs64.8%
42Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs63.9%
43Qwen1.5-14B· Stanford HELM specialized suites data docs62%
44Command R Plus· Stanford HELM specialized suites data docs61.1%
45GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs59.7%
46Llama 2 70b· Stanford HELM specialized suites data docs59.7%
47Claude Haiku 3.5· Stanford HELM specialized suites data docs58.8%
48gemma-7b· Stanford HELM specialized suites data docs57.4%
49Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs57.4%
50Open Mistral Nemo· Stanford HELM specialized suites data docs56.5%

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