Compare AI models

Specifications
Parameters
16B12B
Context window
32k128k
Benchmarks
HumanEval
57.3%72%
MATH
27.9%48.1%
MMLU
55.7%69.2%
Overview
CompanyDeepSeekMistral
Release dateMay 16 2024Sep 11 2024
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

Pixtral 24.09 leads DeepSeek V2 Lite on 3 of the 3 benchmarks they both report (HumanEval, MATH, MMLU). DeepSeek V2 Lite shipped 118 days before Pixtral 24.09, so benchmark comparisons should account for the intervening progress.

DeepSeek V2 Lite has 16B parameters, while Pixtral 24.09 has 12B. Context windows are 32k (DeepSeek V2 Lite) vs 128k (Pixtral 24.09).

On HumanEval, Pixtral 24.09 leads at 72% vs DeepSeek V2 Lite at 57.3%. On MATH, Pixtral 24.09 leads at 48.1% vs DeepSeek V2 Lite at 27.9%. On MMLU, Pixtral 24.09 leads at 69.2% vs DeepSeek V2 Lite at 55.7%.