Skip to content

Benchmark

HELM MMLU: College Physics - EM

Scores are percentages. Higher is better.

Models measured
64
Scores recorded
68
Sources
1
IQ weight
2.02

Score distribution

51.7%
Mean
51%
Median
13.3%
Std dev
36.3%
10th percentile
69.6%
90th percentile
19.6%
Min
19.6%86.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.02 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
1gemini-1.5-pro-002· Stanford HELM specialized suites data docs86.3%
2DeepSeek V3· Stanford HELM specialized suites data docs81.4%
3Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs77.5%
4gemini-1.5-pro-001· Stanford HELM specialized suites data docs74.5%
5Claude 3 Opus· Stanford HELM specialized suites data docs71.6%
6Claude 3.5 Sonnetsetting not named by the source· Stanford HELM specialized suites data docs69.6%
7Gemini 2.0 Flash Experimental· Stanford HELM specialized suites data docs69.6%
8gemini-1.5-flash-001· Stanford HELM specialized suites data docs69.6%
9Llama 3.1 405B Instruct Turbo· Stanford HELM specialized suites data docs69.6%
10GPT 4osetting not named by the source· Stanford HELM specialized suites data docs68.6%
11GPT 4osetting not named by the source· Stanford HELM specialized suites data docs68.6%
12Nova Prov1· Stanford HELM specialized suites data docs64.7%
13gemini-1.5-flash-002· Stanford HELM specialized suites data docs63.7%
14GPT 4· Stanford HELM specialized suites data docs62.7%
15Nova Litev1· Stanford HELM specialized suites data docs60.8%
16Qwen2 72B Instruct· Stanford HELM specialized suites data docs59.8%
17Qwen2.5 72B Instruct Turbo· Stanford HELM specialized suites data docs58.8%
18mixtral-8x22b· Stanford HELM specialized suites data docs56.9%
19Claude 3 Sonnet· Stanford HELM specialized suites data docs55.9%
20GPT-4o mini· Stanford HELM specialized suites data docs55.9%
21Meta Llama 3.1 70B Instruct Turbo· Stanford HELM specialized suites data docs55.9%
22Mistral Large2407· Stanford HELM specialized suites data docs55.9%
23Phi-3 Small 8K Instruct· Stanford HELM specialized suites data docs55.9%
24Qwen1.5-72B· Stanford HELM specialized suites data docs55.9%
25text-unicorn· Stanford HELM specialized suites data docs54.9%
26DBRX-instruct· Stanford HELM specialized suites data docs53.9%
27GPT-4 Turbo· Stanford HELM specialized suites data docs53.9%
28Llama-3.2-90B-Vision-Instruct-Turbo· Stanford HELM specialized suites data docs53.9%
29Meta Llama 3 70b· Stanford HELM specialized suites data docs52.9%
30Phi-3 Medium 4K Instruct· Stanford HELM specialized suites data docs52.9%
31Claude Haiku 3.5· Stanford HELM specialized suites data docs52%
32Command R Plus· Stanford HELM specialized suites data docs52%
33Llama 3.3 70B Instruct Turbo· Stanford HELM specialized suites data docs52%
34Jamba 1.5 Large· Stanford HELM specialized suites data docs51%
35Qwen1.5-110B-Chat· Stanford HELM specialized suites data docs51%
36Qwen2.5 7B Instruct Turbo· Stanford HELM specialized suites data docs51%
37Claude 2.1· Stanford HELM specialized suites data docs50%
38gemma-2-9b· Stanford HELM specialized suites data docs50%
39gemma2-27b· Stanford HELM specialized suites data docs50%
40Nova Microv1· Stanford HELM specialized suites data docs50%
41Yi-34B· Stanford HELM specialized suites data docs50%
42Claude Instant 1.2· Stanford HELM specialized suites data docs49%
43Claude 3 Haiku· Stanford HELM specialized suites data docs48%
44Jamba 1.5 Mini· Stanford HELM specialized suites data docs48%
45Qwen1.5-14B· Stanford HELM specialized suites data docs48%
46GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs47.1%
47Qwen1.5-7B· Stanford HELM specialized suites data docs47.1%
48GPT-3.5 Turbosetting not named by the source· Stanford HELM specialized suites data docs46.1%
49Snowflake Arctic Instruct· Stanford HELM specialized suites data docs46.1%
50Meta Llama 3 8b· Stanford HELM specialized suites data docs45.1%

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