What it does
This page is a collection published by the deepseek-ai organization on Hugging Face for the DeepSeek-V4 family. It lists model repositories you can download, and the page shows download counts in the millions for some of them. The V4 line comes in two sizes, Flash and Pro. DeepSeek-V4-Flash is listed at 291B and DeepSeek-V4-Pro at 1.6T, both as text generation models, each with a Base version. There are also DSpark variants of both (165B for Flash, 1.7T for Pro) and dated updates named V4-Flash-0731 and V4-Pro-0813. V4-Flash-Vision-Exp, V4.1-Flash and V4.1-Flash-1017 are tagged image-text-to-text, and the last two are listed at 763B. The organization page links to many earlier collections, including V3.2, V3.1, R1, V3, OCR, Math, VL2, Janus, Prover and the Coder models. The page does not state licenses, context lengths or benchmark results, so check each model card before relying on a release.
Capabilities
- V4 Flash and Pro models for text generation, each with a Base version
- V4.1-Flash and V4.1-Flash-1017, tagged for image-text-to-text use
- Model sizes shown for each release, from 165B to 1.7T
- Dated updates and experimental variants such as Vision-Exp and DSpark
- Links to earlier DeepSeek collections such as V3.2, V3.1 and R1
When to use it
- You want to download DeepSeek's latest published model weights
- You need to compare the sizes and variants of the V4 line before choosing one
- You want to find an earlier DeepSeek release such as R1 or V3
