Skip to content

Benchmark

HELM MMLU: Global Facts - EM

Factuality benchmarks measure how often models get facts right. Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
1.69

Score distribution

49.8%
Mean
50%
Median
12.2%
Std dev
34%
10th percentile
66%
90th percentile
25%
Min
25%80%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.69 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 docs80%
2gemini-1.5-pro-002· Stanford HELM specialized suites data docs77%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs72%
4Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs71%
5GPT 4osetting not named by the source· Stanford HELM specialized suites data docs69%
6DeepSeek V3· Stanford HELM specialized suites data docs68%
7Claude 3 Opus· Stanford HELM specialized suites data docs66%
8Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs66%
9gemini-1.5-pro-001· Stanford HELM specialized suites data docs66%
10GPT 4osetting not named by the source· Stanford HELM specialized suites data docs64%
11GPT 4· Stanford HELM specialized suites data docs62%
12Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs61%
13Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs61%
14Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs60%
15GPT 4 1106 Preview· Stanford HELM specialized suites data docs58%
16GPT-4 Turbo· Stanford HELM specialized suites data docs58%
17Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs58%
18Qwen2 72B Instruct· Stanford HELM specialized suites data docs58%
19Mistral Large2407· Stanford HELM specialized suites data docs56%
20mixtral-8x22b· Stanford HELM specialized suites data docs56%
21Qwen1.5-72B· Stanford HELM specialized suites data docs56%
22Claude 2.1· Stanford HELM specialized suites data docs55%
23Nova Litev1· Stanford HELM specialized suites data docs55%
24Jamba 1.5 Large· Stanford HELM specialized suites data docs54%
25Nova Prov1· Stanford HELM specialized suites data docs54%
26Claude 3 Sonnet· Stanford HELM specialized suites data docs53%
27gemini-1.5-flash-001· Stanford HELM specialized suites data docs53%
28gemma-2-9b· Stanford HELM specialized suites data docs53%
29text-unicorn· Stanford HELM specialized suites data docs53%
30Yi-34B· Stanford HELM specialized suites data docs53%
31Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs52%
32Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs51%
33Claude Haiku 3.5· Stanford HELM specialized suites data docs50%
34Command R Plus· Stanford HELM specialized suites data docs50%
35Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs50%
36Meta Llama 3 70b· Stanford HELM specialized suites data docs49%
37Qwen1.5-14B· Stanford HELM specialized suites data docs49%
38Claude 3 Haiku· Stanford HELM specialized suites data docs47%
39gemini-1.5-flash-002· Stanford HELM specialized suites data docs47%
40Llama 2 70b· Stanford HELM specialized suites data docs47%
41DBRX-instruct· Stanford HELM specialized suites data docs46%
42DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs46%
43GPT-4o mini· Stanford HELM specialized suites data docs45%
44Mistral Small· Stanford HELM specialized suites data docs45%
45Nova Microv1· Stanford HELM specialized suites data docs44%
46gemma2-27b· Stanford HELM specialized suites data docs43%
47Jamba 1.5 Mini· Stanford HELM specialized suites data docs43%
48Yi-6B· Stanford HELM specialized suites data docs43%
49Command R+· Stanford HELM specialized suites data docs42%
50gemma-7b· Stanford HELM specialized suites data docs42%

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