DeepSeek-V4 Released

·2026.04.24 18:00

Key point

DeepSeek has released the V4 model family, which supports a 1M token context.

1 / 2

DeepSeek-V4 Preview is a MoE language model series with improved efficiency.

V4-Pro: 1.6T params
V4-Flash: 284B params
Both models support a default 1 million token context window.
A new hybrid attention architecture has been applied, significantly reducing compute and memory costs.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.