Compare AI models

Specifications
Parameters
16B3B
Context window
32k128k
Benchmarks
HumanEval
57.3%77.4%
MATH
27.9%51.7%
Overview
CompanyDeepSeekMistral
Release dateMay 16 2024Oct 16 2024
AccessOpen WeightClosed
Model detailsView modelView model

Frequently asked questions

Ministral 3B leads DeepSeek V2 Lite on 2 of the 2 benchmarks they both report (HumanEval, MATH). DeepSeek V2 Lite shipped 153 days before Ministral 3B, so benchmark comparisons should account for the intervening progress.

DeepSeek V2 Lite has 16B parameters, while Ministral 3B has 3B. Context windows are 32k (DeepSeek V2 Lite) vs 128k (Ministral 3B). DeepSeek V2 Lite is open weight, while Ministral 3B is closed.

On HumanEval, Ministral 3B leads at 77.4% vs DeepSeek V2 Lite at 57.3%. On MATH, Ministral 3B leads at 51.7% vs DeepSeek V2 Lite at 27.9%.