Meituan LongCat-2.0-Preview quietly launched: no announcement, no open source

robot
Abstract generation in progress

CryptoWorld News: Meituan has launched a new model, Longcat-2.0-preview, on the Longcat API platform. The update log is dated April 20, but no official announcement or technical report has been released. Previously, Longcat series models were accompanied by official blogs and technical reports, and were open-sourced on Hugging Face and GitHub. The update log for 2.0-preview does not include an open-source link, only available through API, listing three capabilities: targeted at agent development, supporting tool invocation, multi-step reasoning, and long-context tasks, excelling in code generation and complex instruction execution. On April 24, multiple media outlets citing insiders reported that the total parameters of this model exceed one trillion, adopting a MoE architecture, supporting a 1 million context window, with parameter count roughly the same as DeepSeek v4. The entire training and inference process is based on domestic computing clusters, utilizing 50,000 to 60k domestically produced acceleration cards, making it the largest-scale training task completed with domestic computing power to date. During testing, a daily free quota of 10 million tokens was provided.

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