Skip to content

Benchmark

HELM MMLU: Virology - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
2.19

Score distribution

54.5%
Mean
55.4%
Median
4.7%
Std dev
47%
10th percentile
59%
90th percentile
39.2%
Min
39.2%60.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.19 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 docs60.2%
2GPT-4 Turbo· Stanford HELM specialized suites data docs60.2%
3DeepSeek V3· Stanford HELM specialized suites data docs59.6%
4GPT 4· Stanford HELM specialized suites data docs59.6%
5GPT 4osetting not named by the source· Stanford HELM specialized suites data docs59.6%
6mixtral-8x22b· Stanford HELM specialized suites data docs59.6%
7GPT 4 1106 Preview· Stanford HELM specialized suites data docs59%
8Meta Llama 3 70b· Stanford HELM specialized suites data docs59%
9Mistral Large2407· Stanford HELM specialized suites data docs59%
10Nova Prov1· Stanford HELM specialized suites data docs59%
11Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs58.4%
12Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs58.4%
13Qwen1.5-72B· Stanford HELM specialized suites data docs58.4%
14Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs58.4%
15Claude 3 Opus· Stanford HELM specialized suites data docs57.8%
16Claude 3 Sonnet· Stanford HELM specialized suites data docs57.8%
17GPT 4osetting not named by the source· Stanford HELM specialized suites data docs57.8%
18Jamba 1.5 Mini· Stanford HELM specialized suites data docs57.8%
19Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs57.8%
20Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs57.8%
21Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs57.2%
22text-unicorn· Stanford HELM specialized suites data docs57.2%
23Yi-34B· Stanford HELM specialized suites data docs57.2%
24Claude Haiku 3.5· Stanford HELM specialized suites data docs56.6%
25DBRX-instruct· Stanford HELM specialized suites data docs56.6%
26gemini-1.5-flash-001· Stanford HELM specialized suites data docs56.6%
27gemini-1.5-pro-002· Stanford HELM specialized suites data docs56.6%
28Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs56.6%
29Meta Llama 3 8b· Stanford HELM specialized suites data docs56.6%
30Command R Plus· Stanford HELM specialized suites data docs56%
31gemma2-27b· Stanford HELM specialized suites data docs56%
32Open Mistral Nemo· Stanford HELM specialized suites data docs56%
33Qwen2 72B Instruct· Stanford HELM specialized suites data docs56%
34Claude 2.1· Stanford HELM specialized suites data docs55.4%
35DeepSeek LLM 67B Chat· Stanford HELM specialized suites data docs55.4%
36Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs55.4%
37gemini-1.5-pro-001· Stanford HELM specialized suites data docs55.4%
38Jamba 1.5 Large· Stanford HELM specialized suites data docs55.4%
39Mistral Largesetting not named by the source· Stanford HELM specialized suites data docs55.4%
40Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs55.4%
41Claude Instant 1.2· Stanford HELM specialized suites data docs54.8%
42gemma-7b· Stanford HELM specialized suites data docs54.8%
43Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs54.8%
44Claude 3 Haiku· Stanford HELM specialized suites data docs54.2%
45Command R+· Stanford HELM specialized suites data docs54.2%
46GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs54.2%
47Mistral Small· Stanford HELM specialized suites data docs54.2%
48Nova Litev1· Stanford HELM specialized suites data docs54.2%
49Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs54.2%
50GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs53.6%

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