Skip to content

Benchmark

HELM MMLU: Prehistory - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
2.11

Score distribution

82.3%
Mean
85.8%
Median
11%
Std dev
71%
10th percentile
92.6%
90th percentile
31.8%
Min
31.8%95.1%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.11 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 docs95.1%
2Claude 3 Opus· Stanford HELM specialized suites data docs94.1%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs94.1%
4GPT 4osetting not named by the source· Stanford HELM specialized suites data docs93.8%
5GPT 4osetting not named by the source· Stanford HELM specialized suites data docs93.5%
6Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs92.9%
7gemini-1.5-pro-002· Stanford HELM specialized suites data docs92.6%
8GPT 4· Stanford HELM specialized suites data docs92.6%
9Nova Prov1· Stanford HELM specialized suites data docs92.6%
10DeepSeek V3· Stanford HELM specialized suites data docs92.3%
11GPT-4 Turbo· Stanford HELM specialized suites data docs92%
12Mistral Large2407· Stanford HELM specialized suites data docs92%
13GPT 4 1106 Preview· Stanford HELM specialized suites data docs91.7%
14Qwen2 72B Instruct· Stanford HELM specialized suites data docs91.4%
15Meta Llama 3 70b· Stanford HELM specialized suites data docs91%
16Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs91%
17Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs90.4%
18Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs89.8%
19Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs89.5%
20Jamba 1.5 Large· Stanford HELM specialized suites data docs89.2%
21Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs88.6%
22Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs88%
23Qwen1.5-72B· Stanford HELM specialized suites data docs88%
24Claude Haiku 3.5· Stanford HELM specialized suites data docs87.7%
25gemma2-27b· Stanford HELM specialized suites data docs87.7%
26Yi-34B· Stanford HELM specialized suites data docs87.7%
27gemini-1.5-pro-001· Stanford HELM specialized suites data docs87%
28mixtral-8x22b· Stanford HELM specialized suites data docs87%
29text-unicorn· Stanford HELM specialized suites data docs87%
30gemini-1.5-flash-001· Stanford HELM specialized suites data docs86.7%
31Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs86.7%
32Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs86.7%
33Claude 3 Sonnet· Stanford HELM specialized suites data docs86.4%
34DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs85.8%
35Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs85.8%
36Nova Litev1· Stanford HELM specialized suites data docs84.9%
37DBRX-instruct· Stanford HELM specialized suites data docs84%
38Llama 2 70b· Stanford HELM specialized suites data docs84%
39Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs83.6%
40GPT-4o mini· Stanford HELM specialized suites data docs83.3%
41Claude 3 Haiku· Stanford HELM specialized suites data docs82.4%
42Jamba 1.5 Mini· Stanford HELM specialized suites data docs82.4%
43Claude 2.1· Stanford HELM specialized suites data docs82.1%
44Command R Plus· Stanford HELM specialized suites data docs82.1%
45gemma-2-9b· Stanford HELM specialized suites data docs81.2%
46Mistral Small· Stanford HELM specialized suites data docs80.2%
47Jamba Instruct· Stanford HELM specialized suites data docs79.6%
48Snowflake Arctic Instruct· Stanford HELM specialized suites data docs79%
49GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs78.7%
50Nova Microv1· Stanford HELM specialized suites data docs78.7%

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