Skip to content

Benchmark

HELM MMLU: High School Mathematics - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
1.35

Score distribution

43.8%
Mean
44.1%
Median
13.1%
Std dev
29.3%
10th percentile
63%
90th percentile
9.3%
Min
9.3%76.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.35 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
1DeepSeek V3· Stanford HELM specialized suites data docs76.7%
2gemini-1.5-pro-002· Stanford HELM specialized suites data docs73%
3gemini-1.5-pro-001· Stanford HELM specialized suites data docs68.5%
4Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs66.3%
5Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs63.7%
6Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs63.3%
7Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs63%
8Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs63%
9Nova Prov1· Stanford HELM specialized suites data docs61.5%
10Qwen2 72B Instruct· Stanford HELM specialized suites data docs59.6%
11Claude 3 Opus· Stanford HELM specialized suites data docs58.5%
12Nova Litev1· Stanford HELM specialized suites data docs58.1%
13GPT 4· Stanford HELM specialized suites data docs54.4%
14Jamba 1.5 Large· Stanford HELM specialized suites data docs54.4%
15Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs52.6%
16GPT-4 Turbo· Stanford HELM specialized suites data docs51.5%
17gemini-1.5-flash-001· Stanford HELM specialized suites data docs51.1%
18Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs50.4%
19gemini-1.5-flash-002· Stanford HELM specialized suites data docs50%
20Qwen1.5-72B· Stanford HELM specialized suites data docs50%
21Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs49.6%
22Meta Llama 3 70b· Stanford HELM specialized suites data docs49.3%
23text-unicorn· Stanford HELM specialized suites data docs49.3%
24gemma-2-9b· Stanford HELM specialized suites data docs48.5%
25GPT 4osetting not named by the source· Stanford HELM specialized suites data docs48.1%
26GPT 4osetting not named by the source· Stanford HELM specialized suites data docs48.1%
27Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs47.8%
28gemma2-27b· Stanford HELM specialized suites data docs47%
29Yi-34B· Stanford HELM specialized suites data docs46.7%
30mixtral-8x22b· Stanford HELM specialized suites data docs46.3%
31DBRX-instruct· Stanford HELM specialized suites data docs44.4%
32Qwen1.5-14B· Stanford HELM specialized suites data docs44.4%
33Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs44.1%
34Nova Microv1· Stanford HELM specialized suites data docs44.1%
35Claude 3 Sonnet· Stanford HELM specialized suites data docs42.2%
36GPT-4o mini· Stanford HELM specialized suites data docs41.9%
37Meta Llama 3 8b· Stanford HELM specialized suites data docs41.5%
38Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs40.7%
39Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs40.4%
40gemma-7b· Stanford HELM specialized suites data docs40%
41Jamba 1.5 Mini· Stanford HELM specialized suites data docs38.9%
42DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs37.4%
43Claude 3 Haiku· Stanford HELM specialized suites data docs37%
44Command R Plus· Stanford HELM specialized suites data docs37%
45Mistral Large2407· Stanford HELM specialized suites data docs37%
46Qwen1.5-7B· Stanford HELM specialized suites data docs36.7%
47Snowflake Arctic Instruct· Stanford HELM specialized suites data docs36.3%
48Claude Haiku 3.5· Stanford HELM specialized suites data docs35.9%
49Llama 2 70b· Stanford HELM specialized suites data docs35.9%
50Claude 2.1· Stanford HELM specialized suites data docs35.2%

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