DeepSeek-V3

This repository contains the DeepSeek-V3 language model project and related source files.

github.com/deepseek-ai/DeepSeek-V3
DeepSeek-V3 homepage screenshot

About DeepSeek-V3

This GitHub repository hosts the DeepSeek-V3 language model project and related source files. It describes a Mixture-of-Experts language model with 671B total parameters and 37B activated per token, with 128K context length for DeepSeek-V3-Base and DeepSeek-V3. The repository includes inference code, model download links via Hugging Face, and evaluation result tables.

The models were pre-trained on 14.8 trillion tokens, followed by supervised fine-tuning and reinforcement learning stages. Guidance for running the model locally is included in the repository. Multi-token prediction support is noted as under active development.

Key features

  • Mixture-of-Experts model with 671B total parameters, 37B activated per token
  • 128K context length for DeepSeek-V3-Base and DeepSeek-V3
  • Model weights available via Hugging Face download links
  • Inference code and instructions for running the model locally
  • Multi-token prediction module weights included alongside main weights
  • Evaluation tables and training details in repository documentation
Best for: Researchers and developers studying or building with large Mixture-of-Experts language models.
Good for
  • Open source repository with code and model weights
  • Provides DeepSeek-V3-Base and DeepSeek-V3 versions with 128K context
  • Includes inference code and local-run guidance
Watch out for
  • Multi-token prediction support is under active development

DeepSeek-V3 at a glance

Type
Open-source project
Pricing
Not confirmed. Check the site
Open source
Yes · Source code
Language
English
Last checked
Oct 5, 2026 · Online when checked
Listed
Oct 5, 2026

DeepSeek-V3 alternatives

Tools and sites you could use instead, ranked by how closely they match.

All 24 alternatives

Popular in Models & ML Research

Well-known listings from the same topic.

All 69

DeepSeek-V3 FAQ

What versions of DeepSeek-V3 are available for download?
DeepSeek-V3-Base and DeepSeek-V3, each listed with 671B total parameters, 37B activated parameters, and 128K context length, available via Hugging Face.
What makes up the 685B total size on Hugging Face?
671B for the main model weights and 14B for the multi-token prediction module weights.
Where is information about running the model locally?
In the How to Run Locally section and README_WEIGHTS.md, with multiple local-run options from community and hardware partners.
Is something wrong or out of date?

Tell us and an editor will check it.

Own this website?

Claim the listing and suggest updates yourself. An editor checks every claim first.

Claim this listing