---
id: 20260620-T0-01
title: "DeepSeek-V4发布：支持百万Token上下文的高效MoE模型"
title_en: "DeepSeek-V4 Released: Efficient MoE Models with Million-Token Context"
url: https://ai.daily.yangsir.net/daily/20260620-T0-01
issue_date: 2026-06-20
publish_date: 2026-06-19T04:00:00.000Z
category: release
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2606.19348
---

# DeepSeek-V4发布：支持百万Token上下文的高效MoE模型

DeepSeek 发布了 V4 系列预览版，包括两个 MoE 模型：DeepSeek-V4-Pro（1.6T 参数，激活 49B）和 DeepSeek-V4-Flash（284B 参数，激活 13B）。两者均支持百万级 Token 上下文，旨在提供高效的长文本推理能力。

## English Version

**DeepSeek-V4 Released: Efficient MoE Models with Million-Token Context**

DeepSeek released the V4 series preview, featuring two Mixture-of-Experts models: DeepSeek-V4-Pro (1.6T params, 49B activated) and V4-Flash (284B params, 13B activated). Both support million-token context windows for efficient long-context reasoning.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2606.19348)

**详情页**：https://ai.daily.yangsir.net/daily/20260620-T0-01

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*