Qingyang Zhang 张庆阳

News

• 2026.11Hy4 is released, with stronger long-horizon agentic capabilities and native visual understanding.
• 2026.08Hy4 preview (770B-A49B, 1M context) is released and open-sourced. [HF]
• 2026.05Joining Tencent Hunyuan LLM as a research intern (Qingyun Talent Program).
• 2026.02P1-VL is released, bridging visual perception and scientific reasoning in Physics Olympiads. [Paper]
• 2025.12Gave a talk about "LLM RL with Internal Reward" @ BUAA.
• 2025.11P1 is released, the first open-source model to win gold at IPhO 2025. [Project]
• 2025.09One paper about Unsupervised LLM Reasoning Incentivization accepted by NeurIPS as a Spotlight.
• 2025.06Starting an internship at Shanghai AI Lab, mentored by Dr. Ganqu Cui.
• 2025.06Gave a talk about "LLM RL with Internal Reward" @ Huawei 2012 Lab.
• 2025.01One paper about test-time adaptation accepted by ICLR.
• 2024.05One paper about Out-of-distribution Detection accepted by NeurIPS.
• 2024.04Starting an internship at Tencent AI Lab, mentored by Dr. Yatao Bian.
• 2023.04Two papers accepted by ICML, including one Oral. Thanks to all co-authors.

Foundation Models I contributed to

Publications

Right Question is Already Half the Answer: Fully Unsupervised LLM Reasoning Incentivization

Qingyang Zhang, Haitao Wu, Changqing Zhang, Peilin Zhao, Yatao Bian

NeurIPS 2025Spotlight

Incentivize LLM reasoning without any external supervision (human-verified reasoning traces, golden answers or pretrained reward models).

PaperCode145106

COME: Test-time Adaption by Conservatively Minimizing Entropy

Qingyang Zhang, Yatao Bian, Xinke Kong, Peilin Zhao, Changqing Zhang

ICLR 2025

A simple learning principle for test-time adaptation.

PaperCode5024

The Best of Both Worlds: On the Dilemma of Out-of-Distribution Detection

Qingyang Zhang, Qiuxuan Feng, Joey Tianyi Zhou, Yatao Bian, Qinghua Hu, Changqing Zhang

NeurIPS 2024

Resolve the conflict between OOD detection and generalization for dual-optimal performance.

PaperCode1911

Provable Dynamic Fusion for Low-quality Multimodal Learning

Qingyang Zhang, Haitao Wu, Changqing Zhang, Qinghua Hu, Huazhu Fu, Joey Tianyi Zhou, Xi Peng

ICML 2023

Theory-inspired dynamic fusion strategy for quality-varying modalities in the real world.

PaperCode310131

Calibrating Multimodal Learning

Huan Ma, Qingyang Zhang^*, Changqing Zhang, Bingzhe Wu, Huazhu Fu, Joey Tianyi Zhou, Qinghua Hu (*co-first author)

ICML 2023Oral

Mitigate the greedy nature of multimodal learning by regularizing model confidence.

PaperCode9020

Survey

Multimodal Fusion on Low-quality Data: A Comprehensive Survey

Qingyang Zhang, Yake Wei, Zongbo Han, Huazhu Fu, Xi Peng, Cheng Deng, Qinghua Hu, Cai Xu, Jie Wen, Di Hu, Changqing Zhang

arXiv 2024

A systematic survey on fusion of low-quality multimodal data.

PaperAwesome List26653

Internships

• 
  Tencent Hunyuan LLM · Post-train Team · Qingyun Talent Program2026.05 – Present

  LLM post-training (Hy4 preview, Hy4).

• 
  Alibaba Tongyi Qwen2026.03 – 2026.05

  Mentor: Dr. Minghao Wu.

• 
  Shanghai AI Lab · P1 Team2025.06 – 2026.03

  Mentor: Dr. Ganqu Cui. RL post-training for physics reasoning (P1, P1-VL).

• 
  Tencent AI Lab · Machine Learning Group2024.04 – 2025.05

  Mentors: Dr. Yatao Bian, Dr. Peilin Zhao. Focus on foundation models.