Skip to content

Benchmark

HELM MMLU: Anatomy - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
1.99

Score distribution

70.4%
Mean
72.6%
Median
12.5%
Std dev
53.3%
10th percentile
82.2%
90th percentile
22.2%
Min
22.2%91.1%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.99 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 4osetting not named by the source· Stanford HELM specialized suites data docs91.1%
2GPT 4osetting not named by the source· Stanford HELM specialized suites data docs91.1%
3DeepSeek V3· Stanford HELM specialized suites data docs86.7%
4Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs85.9%
5Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs84.4%
6gemini-1.5-pro-001· Stanford HELM specialized suites data docs83%
7gemini-1.5-pro-002· Stanford HELM specialized suites data docs83%
8GPT-4 Turbo· Stanford HELM specialized suites data docs82.2%
9Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs82.2%
10Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs82.2%
11Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs80.7%
12GPT 4 1106 Preview· Stanford HELM specialized suites data docs80.7%
13Nova Prov1· Stanford HELM specialized suites data docs80.7%
14Claude 3 Opus· Stanford HELM specialized suites data docs80%
15gemini-1.5-flash-001· Stanford HELM specialized suites data docs80%
16GPT 4· Stanford HELM specialized suites data docs80%
17Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs80%
18Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs80%
19Claude Haiku 3.5· Stanford HELM specialized suites data docs79.3%
20gemini-1.5-flash-002· Stanford HELM specialized suites data docs79.3%
21Jamba 1.5 Large· Stanford HELM specialized suites data docs79.3%
22Qwen2 72B Instruct· Stanford HELM specialized suites data docs79.3%
23Meta Llama 3 70b· Stanford HELM specialized suites data docs78.5%
24Mistral Large2407· Stanford HELM specialized suites data docs78.5%
25Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs77.8%
26gemma2-27b· Stanford HELM specialized suites data docs77%
27GPT-4o mini· Stanford HELM specialized suites data docs77%
28Yi-34B· Stanford HELM specialized suites data docs74.8%
29mixtral-8x22b· Stanford HELM specialized suites data docs74.1%
30Qwen1.5-72B· Stanford HELM specialized suites data docs73.3%
31text-unicorn· Stanford HELM specialized suites data docs73.3%
32Claude 2.1· Stanford HELM specialized suites data docs72.6%
33Nova Microv1· Stanford HELM specialized suites data docs72.6%
34Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs72.6%
35Nova Litev1· Stanford HELM specialized suites data docs71.9%
36Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs71.9%
37Claude 3 Haiku· Stanford HELM specialized suites data docs71.1%
38Claude 3 Sonnet· Stanford HELM specialized suites data docs71.1%
39Jamba 1.5 Mini· Stanford HELM specialized suites data docs71.1%
40gemma-2-9b· Stanford HELM specialized suites data docs70.4%
41GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs69.6%
42Meta Llama 3 8b· Stanford HELM specialized suites data docs69.6%
43Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs69.6%
44Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs68.9%
45Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs67.4%
46Mistral Small· Stanford HELM specialized suites data docs67.4%
47DBRX-instruct· Stanford HELM specialized suites data docs66.7%
48DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs66.7%
49GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs65.9%
50Snowflake Arctic Instruct· Stanford HELM specialized suites data docs65.2%

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