Skip to content

Benchmark

HELM MMLU: Astronomy - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
2.17

Score distribution

81.8%
Mean
84.9%
Median
12.7%
Std dev
67.1%
10th percentile
93.4%
90th percentile
34.2%
Min
34.2%97.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.17 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 docs97.4%
2Claude 3 Opus· Stanford HELM specialized suites data docs96.7%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs96.1%
4GPT 4osetting not named by the source· Stanford HELM specialized suites data docs94.7%
5GPT 4 1106 Preview· Stanford HELM specialized suites data docs94.1%
6GPT 4osetting not named by the source· Stanford HELM specialized suites data docs94.1%
7GPT-4 Turbo· Stanford HELM specialized suites data docs94.1%
8gemini-1.5-pro-002· Stanford HELM specialized suites data docs93.4%
9GPT 4· Stanford HELM specialized suites data docs93.4%
10Qwen2 72B Instruct· Stanford HELM specialized suites data docs93.4%
11Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs93.4%
12Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs92.8%
13DeepSeek V3· Stanford HELM specialized suites data docs92.1%
14Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs92.1%
15Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs92.1%
16Meta Llama 3 70b· Stanford HELM specialized suites data docs92.1%
17Mistral Large2407· Stanford HELM specialized suites data docs92.1%
18gemini-1.5-pro-001· Stanford HELM specialized suites data docs91.4%
19Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs90.8%
20Claude 3 Haiku· Stanford HELM specialized suites data docs90.1%
21Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs90.1%
22Yi-34B· Stanford HELM specialized suites data docs90.1%
23gemini-1.5-flash-002· Stanford HELM specialized suites data docs89.5%
24Nova Prov1· Stanford HELM specialized suites data docs89.5%
25Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs88.8%
26gemini-1.5-flash-001· Stanford HELM specialized suites data docs88.2%
27Jamba 1.5 Large· Stanford HELM specialized suites data docs88.2%
28mixtral-8x22b· Stanford HELM specialized suites data docs88.2%
29Qwen1.5-72B· Stanford HELM specialized suites data docs86.8%
30Nova Litev1· Stanford HELM specialized suites data docs86.2%
31text-unicorn· Stanford HELM specialized suites data docs86.2%
32Claude 2.1· Stanford HELM specialized suites data docs85.5%
33Claude 3 Sonnet· Stanford HELM specialized suites data docs85.5%
34GPT-4o mini· Stanford HELM specialized suites data docs84.9%
35Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs84.9%
36Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs84.9%
37Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs84.2%
38DBRX-instruct· Stanford HELM specialized suites data docs83.6%
39Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs83.6%
40Claude Haiku 3.5· Stanford HELM specialized suites data docs82.9%
41gemma2-27b· Stanford HELM specialized suites data docs82.9%
42Llama 2 70b· Stanford HELM specialized suites data docs82.9%
43DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs82.2%
44Jamba 1.5 Mini· Stanford HELM specialized suites data docs82.2%
45Nova Microv1· Stanford HELM specialized suites data docs82.2%
46gemma-2-9b· Stanford HELM specialized suites data docs78.9%
47Command R Plus· Stanford HELM specialized suites data docs78.3%
48Mistral Small· Stanford HELM specialized suites data docs77%
49GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs76.3%
50Snowflake Arctic Instruct· Stanford HELM specialized suites data docs76.3%

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