Compare AI models

Specifications
Parameters
16B7B
Context window
4k32k
Benchmarks
HumanEval
45.7%38.4%
MATH
15.2%13.2%
Overview
CompanyDeepSeekMistral
Release dateJan 11 2024May 22 2024
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

DeepSeek-MoE leads Mistral 7B v0.3 on 2 of the 2 benchmarks they both report (HumanEval, MATH). DeepSeek-MoE shipped 132 days before Mistral 7B v0.3, so benchmark comparisons should account for the intervening progress.

DeepSeek-MoE has 16B parameters, while Mistral 7B v0.3 has 7B. Context windows are 4k (DeepSeek-MoE) vs 32k (Mistral 7B v0.3).

On HumanEval, DeepSeek-MoE leads at 45.7% vs Mistral 7B v0.3 at 38.4%. On MATH, DeepSeek-MoE leads at 15.2% vs Mistral 7B v0.3 at 13.2%.