Skip to content

Benchmark

HELM MMLU: Us Foreign Policy - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
2.00

Score distribution

89%
Mean
92%
Median
9.8%
Std dev
79%
10th percentile
96%
90th percentile
26%
Min
26%96%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.00 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 Opus· Stanford HELM specialized suites data docs96%
2Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs96%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs96%
4GPT 4 1106 Preview· Stanford HELM specialized suites data docs96%
5GPT 4osetting not named by the source· Stanford HELM specialized suites data docs96%
6GPT-4 Turbo· Stanford HELM specialized suites data docs96%
7Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs96%
8text-unicorn· Stanford HELM specialized suites data docs96%
9Claude 3 Haiku· Stanford HELM specialized suites data docs95%
10gemma2-27b· Stanford HELM specialized suites data docs95%
11GPT 4· Stanford HELM specialized suites data docs95%
12GPT 4osetting not named by the source· Stanford HELM specialized suites data docs95%
13mixtral-8x22b· Stanford HELM specialized suites data docs95%
14Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs95%
15Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs95%
16Claude 3 Sonnet· Stanford HELM specialized suites data docs94%
17Claude Haiku 3.5· Stanford HELM specialized suites data docs94%
18gemini-1.5-pro-002· Stanford HELM specialized suites data docs94%
19Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs94%
20Meta Llama 3 70b· Stanford HELM specialized suites data docs94%
21Qwen1.5-72B· Stanford HELM specialized suites data docs94%
22Qwen2 72B Instruct· Stanford HELM specialized suites data docs94%
23DBRX-instruct· Stanford HELM specialized suites data docs93%
24gemini-1.5-flash-001· Stanford HELM specialized suites data docs93%
25gemini-1.5-pro-001· Stanford HELM specialized suites data docs93%
26Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs93%
27Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs93%
28Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs93%
29Nova Prov1· Stanford HELM specialized suites data docs93%
30Claude 2.1· Stanford HELM specialized suites data docs92%
31DeepSeek V3· Stanford HELM specialized suites data docs92%
32Jamba 1.5 Large· Stanford HELM specialized suites data docs92%
33Llama 2 70b· Stanford HELM specialized suites data docs92%
34Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs92%
35Nova Litev1· Stanford HELM specialized suites data docs92%
36DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs91%
37GPT-4o mini· Stanford HELM specialized suites data docs91%
38Jamba Instruct· Stanford HELM specialized suites data docs91%
39Yi-34B· Stanford HELM specialized suites data docs91%
40Claude Instant 1.2· Stanford HELM specialized suites data docs90%
41gemma-2-9b· Stanford HELM specialized suites data docs90%
42Jamba 1.5 Mini· Stanford HELM specialized suites data docs90%
43Mistral Large2407· Stanford HELM specialized suites data docs90%
44Nova Microv1· Stanford HELM specialized suites data docs90%
45Command R Plus· Stanford HELM specialized suites data docs89%
46GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs89%
47Mistral Small· Stanford HELM specialized suites data docs89%
48Open Mistral Nemo· Stanford HELM specialized suites data docs89%
49GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs88%
50Meta Llama 3 8b· Stanford HELM specialized suites data docs88%

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