ByteDance’s Seed Team Open-Sources Seed-OSS-36B Model with 512k Context Window
ByteDance’s Seed team has open-sourced the Seed-OSS model series, which includes 36B-parameter Base and Instruct versions supporting a 512k context window—the longest among open-source models to date. The models were trained on 12 trillion tokens and demonstrated strong performance across multiple benchmarks, achieving 91.7% on AIME24. Their inference budgeting feature allows users to flexibly adjust reasoning length, improving efficiency.
© Copyright Notice
The copyright of the article belongs to the author. Please do not reprint without permission.
Related Posts
No comments yet...