On the first day of DeepSeek V4's release, reactions from Chinese and American developers: architectural innovation is recognized, but the consensus is that it "still lags behind the cutting edge by half a year."

ME News news, April 24 (UTC+8), according to Dongcha Beating monitoring, after the release of DeepSeek V4, the developer community in China and the US has shown a coexistence of "technical recognition and gap consensus." On the US side, Replit CEO Amjad Masad praised V4's attention compression and long-context efficiency improvements as genuine architectural innovations. Developers on Reddit and Hacker News responded positively to the 1M context open-source model and MIT license, but most adopted a wait-and-see attitude of "test first, then talk." CFR researcher Chris McGuire noted that the V4 report itself admits to being 3 to 6 months behind frontier models. On the Chinese side, V2EX and Zhihu communities focused on V4's programming agent capabilities and low-cost strategy. Early feedback suggests that the Pro version is close to the level of Claude Opus 4.6, but there is still a gap in complex deep reasoning tasks. There is high interest in adaptation to Huawei Ascend. In the months before the release, the community had accumulated skepticism due to repeated delays. After the official release, it entered a phase of intensive testing. A common concern for both sides is that V4 performs strongly in coding and long-context scenarios, but has not yet caught up with US frontier closed-source models in comprehensive reasoning. (Source: BlockBeats)
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned