Seed1.5-Thinking: Advancing Superb Reasoning Models with Reinforcement Learning
arXiv:2504.13914, Technical Report
Abstract
Seed1.5-Thinking is a mixture-of-experts reasoning model developed by ByteDance Seed, with 20 billion active parameters out of 200 billion total. The report describes data preparation, reward modeling, reinforcement learning algorithms, and training infrastructure for improving mathematical, scientific, coding, and general-purpose reasoning.
Tao Sun is a contributing author on this ByteDance Seed team report. See the complete author list on arXiv.
Originally released as Seed-Thinking-v1.5; the revised report is titled Seed1.5-Thinking.
BibTeX
@misc{bytedanceseed2025seedthinking,
title = {Seed1.5-Thinking: Advancing Superb Reasoning Models with Reinforcement Learning},
author = {{ByteDance Seed}},
year = {2025},
eprint = {2504.13914},
archivePrefix = {arXiv},
primaryClass = {cs.CL},
doi = {10.48550/arXiv.2504.13914},
url = {https://arxiv.org/abs/2504.13914}
}