Compare AI models

Specifications
Parameters
16B141B
Context window
32k64k
API pricing
Cheapest input
—$2.00Mistral
Cheapest output
—$6.00Mistral
—$700Mistral · zdr
Benchmarks
HumanEval
57.3%45.1%
MMLU
55.7%77.75%
GSM8K
72%78.6%
Overview
CompanyDeepSeekMistral
Release dateMay 16 2024Apr 10 2024
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

Mixtral 8×22B leads DeepSeek V2 Lite on 2 of the 3 benchmarks they both report (HumanEval, MMLU, GSM8K). Mixtral 8×22B shipped 36 days before DeepSeek V2 Lite, so benchmark comparisons should account for the intervening progress.

DeepSeek V2 Lite has 16B parameters, while Mixtral 8×22B has 141B. Context windows are 32k (DeepSeek V2 Lite) vs 64k (Mixtral 8×22B).

On HumanEval, DeepSeek V2 Lite leads at 57.3% vs Mixtral 8×22B at 45.1%. On MMLU, Mixtral 8×22B leads at 77.75% vs DeepSeek V2 Lite at 55.7%. On GSM8K, Mixtral 8×22B leads at 78.6% vs DeepSeek V2 Lite at 72%.