Paper Detail
Hongcheng Gao, Weiheng Li, Yipeng Wei, Yu Lei, Shuo Sun, Yi Li, Yajiao Wang, Jie Luo, Jianwei Wen, Xinyi Zhou, Hongda Li, Yibing Zhang, Henghui Sun, Yufan Zhou, Tianfang Hao, Naihao Xue, Hailong Qu, Jingjing Zhou, Peng Xu, Yongzheng Wei, Yichen Zhu, Ziying Wang, Yibo Wang, Jianshu Zeng, Yuhang Lei, Zeqiang Wang
We introduce Micius-Flash, a 27-billion-parameter model for engineering with native multimodal input, deep reasoning and a 256K-token context window. We design the Micius data pipeline as a knowledge-tree-driven data flywheel comprising three stages: knowledge tree construction and expansion, data synthesis, and quality validation with hard-example augmentation. From this data, we train expert models for semiconductor knowledge and reasoning, hardware coding, and semiconductor multimodal understanding, along with agentic experts. Subsequently, we employ Multi-Teacher On-Policy Distillation (MOPD) to supervise training on sequences generated by Micius itself, thereby integrating the capabilities of these expert models into a unified model. On more than 20 datasets with about 15K records, Micius-Flash averages 64.2% on hardware code benchmarks, 15.0 points above a same-size baseline. It scores 62.4% on a semiconductor-specific agent benchmark, exceeding the same-size baseline by 17.2 points. We further develop Micius Harness for industrial applications, connecting the model to specialized tools, including EDA tools and KiCad, as well as execution environments. Case studies in PCB design, chip design workflow automation, and software development demonstrate the model's ability to use tool execution feedback for multi-step planning, code debugging, and iterative verification. We release Micius as an open-source model to advance the development of agents for science and engineering.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{gao2026micius,
title = {Micius-Flash: Lightweight Agentic Intelligence for Engineering},
author = {Hongcheng Gao and Weiheng Li and Yipeng Wei and Yu Lei and Shuo Sun and Yi Li and Yajiao Wang and Jie Luo and Jianwei Wen and Xinyi Zhou and Hongda Li and Yibing Zhang and Henghui Sun and Yufan Zhou and Tianfang Hao and Naihao Xue and Hailong Qu and Jingjing Zhou and Peng Xu and Yongzheng Wei and Yichen Zhu and Ziying Wang and Yibo Wang and Jianshu Zeng and Yuhang Lei and Zeqiang Wang},
year = {2026},
abstract = {We introduce Micius-Flash, a 27-billion-parameter model for engineering with native multimodal input, deep reasoning and a 256K-token context window. We design the Micius data pipeline as a knowledge-tree-driven data flywheel comprising three stages: knowledge tree construction and expansion, data synthesis, and quality validation with hard-example augmentation. From this data, we train expert models for semiconductor knowledge and reasoning, hardware coding, and semiconductor multimodal underst},
url = {https://huggingface.co/papers/2611.10015},
keywords = {native multimodal input, deep reasoning, 256K-token context window, knowledge-tree-driven data flywheel, data synthesis, quality validation, hard-example augmentation, expert models, semiconductor multimodal understanding, agentic experts, Multi-Teacher On-Policy Distillation, on-policy distillation, unified model, tool execution feedback, multi-step planning, iterative verification, huggingface daily},
eprint = {2611.10015},
archiveprefix = {arXiv},
}
{}