Skip to content

DeepSeek V3-0324

by DeepSeek · deepseek-v3 family

671B

parameters

text-generation code-generation reasoning multilingual math tool-use creative-writing summarization

DeepSeek V3-0324 is the March 2025 update to DeepSeek V3, bringing improved reasoning, coding, and instruction-following capabilities while maintaining the same 671B MoE architecture. This refresh delivers measurable gains on key benchmarks without changing the model's parameter count or context length. As a drop-in upgrade to the original V3, this version benefits from additional training and alignment improvements. It requires the same hardware resources as V3 but delivers stronger performance across the board.

Quick Start with Ollama

ollama run 0324-q4_K_M
Resources Ollama Hugging Face
Creator DeepSeek
Parameters 671B
Architecture mixture-of-experts
Context 128K tokens
Released Mar 24, 2025
License DeepSeek License
Ollama deepseek-v3:0324

Quantization Options

Format File Size VRAM Required Quality Ollama Tag
Q4_K_M rec 350 GB 362 GB 0324-q4_K_M
Q8_0 670 GB 685 GB 0324-q8_0

Compatible Hardware

Q4_K_M requires 362 GB VRAM

Compatible Hardware

HardwareVRAMTypeFitEst. Speed
Mac Studio M4 Ultra 512GB512 GBmacRuns~2 tok/s
106 hardware device(s) cannot run this model at Q4_K_M.

Benchmark Scores

89.0
mmlu