Skip to content

Benchmark

HELM MMLU: College Computer Science - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
1.83

Score distribution

63.1%
Mean
65%
Median
12.3%
Std dev
46%
10th percentile
79%
90th percentile
35%
Min
35%84%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.83 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 docs84%
2Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs83%
3gemini-1.5-pro-002· Stanford HELM specialized suites data docs82%
4Claude 3 Opus· Stanford HELM specialized suites data docs81%
5DeepSeek V3· Stanford HELM specialized suites data docs80%
6Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs79%
7gemini-1.5-pro-001· Stanford HELM specialized suites data docs79%
8GPT 4osetting not named by the source· Stanford HELM specialized suites data docs79%
9Nova Prov1· Stanford HELM specialized suites data docs79%
10Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs78%
11GPT 4osetting not named by the source· Stanford HELM specialized suites data docs77%
12Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs77%
13GPT 4· Stanford HELM specialized suites data docs76%
14Mistral Large2407· Stanford HELM specialized suites data docs75%
15Qwen2 72B Instruct· Stanford HELM specialized suites data docs74%
16Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs73%
17GPT-4 Turbo· Stanford HELM specialized suites data docs72%
18Jamba 1.5 Large· Stanford HELM specialized suites data docs71%
19Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs71%
20gemini-1.5-flash-002· Stanford HELM specialized suites data docs70%
21Meta Llama 3 70b· Stanford HELM specialized suites data docs70%
22mixtral-8x22b· Stanford HELM specialized suites data docs70%
23Nova Litev1· Stanford HELM specialized suites data docs70%
24Claude 2.1· Stanford HELM specialized suites data docs69%
25GPT 4 1106 Preview· Stanford HELM specialized suites data docs69%
26Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs68%
27Qwen1.5-72B· Stanford HELM specialized suites data docs68%
28DBRX-instruct· Stanford HELM specialized suites data docs67%
29GPT-4o mini· Stanford HELM specialized suites data docs67%
30Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs67%
31Claude 3 Haiku· Stanford HELM specialized suites data docs66%
32Claude 3 Sonnet· Stanford HELM specialized suites data docs66%
33gemini-1.5-flash-001· Stanford HELM specialized suites data docs66%
34text-unicorn· Stanford HELM specialized suites data docs65%
35Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs64%
36Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs64%
37Yi-34B· Stanford HELM specialized suites data docs63%
38Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs62%
39gemma2-27b· Stanford HELM specialized suites data docs61%
40Jamba 1.5 Mini· Stanford HELM specialized suites data docs61%
41Claude Instant 1.2· Stanford HELM specialized suites data docs60%
42Command R Plus· Stanford HELM specialized suites data docs59%
43Llama 2 70b· Stanford HELM specialized suites data docs59%
44Nova Microv1· Stanford HELM specialized suites data docs59%
45GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs58%
46GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs58%
47Meta Llama 3 8b· Stanford HELM specialized suites data docs58%
48Claude Haiku 3.5· Stanford HELM specialized suites data docs57%
49Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs57%
50Snowflake Arctic Instruct· Stanford HELM specialized suites data docs57%

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