Paper Detail
Jian Yang, Wei Zhang, Shawn Guo, Zhengmao Ye, Lin Jing, Shark Liu, Yizhi Li, Jiajun Wu, Cening Liu, X. Ma, Yuyang Song, Siwei Wu, Yuwen Li, L. Liao, T. Zheng, Ziling Huang, Zelong Huang, Che Liu, Yan Xing, Renyuan Li, Qingsong Cai, Hanxu Yan, Siyue Wang, Shikai Li, Jason Klein Liu, An Huang, Yongsheng Kang, Jinxing Zhang, Chuan Hao, Haowen Wang, Weicheng Gu, Ran Tao, Mingjie Tang, Peihao Wu, Jianzhou Wang, Xianglong Liu, Weifeng Lv, Bryan Dai
In this report, we introduce the IQuest-Coder-V1 series-(7B/14B/40B/40B-Loop), a new family of code large language models (LLMs). Moving beyond static code representations, we propose the code-flow multi-stage training paradigm, which captures the dynamic evolution of software logic through different phases of the pipeline. Our models are developed through the evolutionary pipeline, starting with the initial pre-training consisting of code facts, repository, and completion data. Following that, we implement a specialized mid-training stage that integrates reasoning and agentic trajectories in 32k-context and repository-scale in 128k-context to forge deep logical foundations. The models are then finalized with post-training of specialized coding capabilities, which is bifurcated into two specialized paths: the thinking path (utilizing reasoning-driven RL) and the instruct path (optimized for general assistance). IQuest-Coder-V1 achieves state-of-the-art performance among competitive models across critical dimensions of code intelligence: agentic software engineering, competitive programming, and complex tool use. To address deployment constraints, the IQuest-Coder-V1-Loop variant introduces a recurrent mechanism designed to optimize the trade-off between model capacity and deployment footprint, offering an architecturally enhanced path for efficacy-efficiency trade-off. We believe the release of the IQuest-Coder-V1 series, including the complete white-box chain of checkpoints from pre-training bases to the final thinking and instruction models, will advance research in autonomous code intelligence and real-world agentic systems.
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{yang2026iquest,
title = {IQuest-Coder-V1 Technical Report},
author = {Jian Yang and Wei Zhang and Shawn Guo and Zhengmao Ye and Lin Jing and Shark Liu and Yizhi Li and Jiajun Wu and Cening Liu and X. Ma and Yuyang Song and Siwei Wu and Yuwen Li and L. Liao and T. Zheng and Ziling Huang and Zelong Huang and Che Liu and Yan Xing and Renyuan Li and Qingsong Cai and Hanxu Yan and Siyue Wang and Shikai Li and Jason Klein Liu and An Huang and Yongsheng Kang and Jinxing Zhang and Chuan Hao and Haowen Wang and Weicheng Gu and Ran Tao and Mingjie Tang and Peihao Wu and Jianzhou Wang and Xianglong Liu and Weifeng Lv and Bryan Dai},
year = {2026},
abstract = {In this report, we introduce the IQuest-Coder-V1 series-(7B/14B/40B/40B-Loop), a new family of code large language models (LLMs). Moving beyond static code representations, we propose the code-flow multi-stage training paradigm, which captures the dynamic evolution of software logic through different phases of the pipeline. Our models are developed through the evolutionary pipeline, starting with the initial pre-training consisting of code facts, repository, and completion data. Following that, },
url = {https://huggingface.co/papers/2603.16733},
keywords = {code large language models, code-flow multi-stage training paradigm, software logic, pre-training, mid-training stage, reasoning, agentic trajectories, 32k-context, 128k-context, post-training, specialized coding capabilities, thinking path, instruct path, reasoning-driven RL, agentic software engineering, competitive programming, complex tool use, recurrent mechanism, deployment footprint, white-box chain of checkpoints, code available, huggingface daily},
eprint = {2603.16733},
archiveprefix = {arXiv},
}
{}