Compare AI models

Specifications
Parameters
16B7B/13B/33B/65B
Context window
32k2k
Benchmarks
HumanEval
57.3%23.7%
MATH
27.9%10.6%
MMLU
55.7%63.4%
GSM8K
72%50.9%
Overview
CompanyDeepSeekMeta
Release dateMay 16 2024Feb 24 2023
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

DeepSeek V2 Lite leads LLaMA 1 on 3 of the 4 benchmarks they both report (HumanEval, MATH, MMLU, GSM8K). LLaMA 1 shipped 447 days before DeepSeek V2 Lite, so benchmark comparisons should account for the intervening progress.

DeepSeek V2 Lite has 16B parameters, while LLaMA 1 has 7B/13B/33B/65B. Context windows are 32k (DeepSeek V2 Lite) vs 2k (LLaMA 1).

On HumanEval, DeepSeek V2 Lite leads at 57.3% vs LLaMA 1 at 23.7%. On MATH, DeepSeek V2 Lite leads at 27.9% vs LLaMA 1 at 10.6%. On MMLU, LLaMA 1 leads at 63.4% vs DeepSeek V2 Lite at 55.7%. On GSM8K, DeepSeek V2 Lite leads at 72% vs LLaMA 1 at 50.9%.