Particle.news
Download on the App Store

DeepSeek Launches V4 Pre-Release With 1 Million-Character Context

The move raises stakes in the U.S.–China AI race following a White House warning on large-scale copying.

Overview

  • DeepSeek, which posted an open-source pre-release Friday, introduced two tiers called V4-Pro and the lower-cost V4-Flash.
  • The company says V4 handles a one million character context in a single pass, letting it answer on hundreds of pages without losing place.
  • DeepSeek claims it made long-context work faster and cheaper than before, which could cut compute costs for tasks like large document review.
  • The firm presents V4 as multimodal and says it runs on Huawei’s latest chips, a notable option as China lacks access to top Nvidia parts.
  • The launch followed OpenAI’s GPT-5.5 reveal hours earlier, and independent tests of V4’s speed, cost, and output quality have not yet appeared.