Release dates
— DeepSeek released public previews and open weights for V4 Pro and Flash together. ↗
DeepSeek-V4-Pro-Preview
The high-capability DeepSeek V4 preview: a 1.6T-total, 49B-active MoE with a 1M context window.
MODEL NUMBER027
- First public date
- 2026-04-24
- Organization
- DeepSeek
- Family
- DeepSeek V / R
01
Specifications
Only published specifications are shown.
- Total parameters
- 1.6T
- Active parameters
- 49B
- Architecture
- MoE
- Context
- 1,000,000 tokens
- Input modalities
- Text
- Open weights
- Yes
- License
- DeepSeek Model License
- Status
- Legacy
02
Overview
Why it mattered
It introduced V4’s sparse-attention, million-token architecture and released the weights alongside the preview.
PreviousDeepSeek-V3.2671B · 128K ctx
CurrentDeepSeek-V4-Pro-Preview1.6T · 1M ctx
04
Architecture & capabilities
Attention
DSA
Model capabilities
AgentCoding
API features
Adjustable reasoning
05
Release history
DeepSeek released public previews and open weights for V4 Pro and Flash together.
View source06
Related models
Select a card to view the related model.
07
Sources
Suggest a correction
Copy this template, add the field, proposed value, and primary source, and share it with the archive maintainer.