Agentic and Reasoning Models

Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent

Technical Report, 2026-06 (key contribution to this project)

Lei Bai^† , Zongsheng Cao , Yang Chen , Zhiyao Cui ... Bowen Zhou , Yuhao Zhou

A 35B agentic model reaches stronger long-horizon reasoning performance through agent-level scaling.

agentic reasoning model long-horizon tasks agentic scaling

Agentic and Reasoning Models

Agents-K1: Towards Agent-native Knowledge Orchestration

Technical Report, 2026-06

Zongsheng Cao , Bihao Zhan , Jinxin Shi , Jiong Wang ... Bo Zhang^† , Lei Bai^†

An end-to-end knowledge orchestration pipeline that converts raw documents into agent-native scientific knowledge graphs.

knowledge orchestration Scholar-KG scientific reasoning

Agentic and Reasoning Models

P1-VL: Bridging Visual Perception and Scientific Reasoning in Physics Olympiads

ICML AI4Math Workshop, 2026 (key contribution to this project)

Yun Luo^*† , Futing Wang^* , Qianjia Cheng^* , Fangchen Yu^* ... Peng Ye^† , Ganqu Cui^†

A family of open-source vision-language models engineered for advanced scientific reasoning.

visual scientific reasoning physics olympiad reinforcement learning

Agentic and Reasoning Models

P1: Mastering Physics Olympiads with Reinforcement Learning

Technical Report, 2025-11 (key contribution to this project)

Jiacheng Chen^* , Qianjia Cheng^* , Fangchen Yu^* , Haiyuan Wan ... Peng Ye^† , Ganqu Cui^†

A family of open-source physics reasoning models trained entirely through reinforcement learning.

physics reasoning model reinforcement learning physics olympiad

Agentic and Reasoning Models

SCI-Verifier: Scientific Verifier with Thinking

ICLR, 2026

Shenghe Zheng^* , Chenyu Huang^* , Fangchen Yu , Junchi Yao ... Ganqu Cui , Peng Ye^†

A unified reasoning-augmented verifier for scientific domains with enhanced reliability.

scientific verifier cross-disciplinary verification verification reliability

Agentic and Reasoning Models

Small Model as Master Orchestrator: Learning Unified Agent-Tool Orchestration with Parallel Subtask Decomposition

arXiv, 2026-04

Wenzhen Yuan^* , Wutao Xiong^* , Fangchen Yu , Shengji Tang ... Wanli Ouyang , Lei Bai

A 4B orchestrator model learns parallel decomposition and tool coordination for agent workflows.

orchestration parallel decomposition tool agents

Agentic Systems

InternAgent-1.5: A Unified Agentic Framework for Long-Horizon Autonomous Scientific Discovery

Technical Report, 2026-02

Shiyang Feng , Runmin Ma , Xiangchao Yan , Yue Fan ... Bo Zhang^† , Lei Bai^†

A unified agentic framework designed for end-to-end scientific discovery in long-horizon tasks.

agentic framework scientific discovery long-horizon task

Agentic Systems

PhysicsMinions: Winning Gold Medals in the Latest Physics Olympiads with a Coevolutionary Multimodal Multi-agent System

arXiv, 2025-09

Fangchen Yu^* , Junchi Yao^* , Ziyi Wang , Haiyuan Wan ... Wanli Ouyang , Peng Ye^†

A coevolutionary multimodal multi-agent system wins gold medals in the latest physics Olympiads.

multi-agent system physics olympiad multimodal physics reasoning

Frontier Benchmarks

HiPhO: How Far Are MLLMs from Humans in the Latest High School Physics Olympiad Benchmark?

ICML, 2026

Fangchen Yu^* , Haiyuan Wan^* , Qianjia Cheng^* , Yuchen Zhang ... Ganqu Cui^† , Peng Ye^†

The first benchmark dedicated to high school physics Olympiads with human-aligned evaluation.

physics olympiad benchmark multimodal physics reasoning human-level comparison

Frontier Benchmarks

ResearchClawBench: A Benchmark for End-to-End Autonomous Scientific Research

arXiv, 2026-05

Wanghan Xu^* , Shuo Li^* , Tianlin Ye , Qinglong Cao ... Lei Bai^† , Wenlong Zhang^†

A benchmark evaluates end-to-end autonomous systems on realistic scientific research tasks.

agentic benchmark autonomous scientific research scientific re-discovery

Frontier Benchmarks

Probing Scientific General Intelligence of LLMs with Scientist-Aligned Workflows

arXiv, 2025-12

Wanghan Xu , Yuhao Zhou , Yifan Zhou , Qinglong Cao ... Wenlong Zhang^† , Lei Bai^†

A scientific intelligence benchmark for LLMs via scientist-aligned workflows.

scientific general intelligence science benchmark LLM evaluation

Frontier Benchmarks

DocReal: Robust Document Dewarping of Real-Life Images via Attention-Enhanced Control Point Prediction

WACV, 2024

Fangchen Yu , Yina Xie , Lei Wu , Yafei Wen ... Jianfeng Mao , Wenye Li^†

An attention-enhanced control point approach improves robust dewarping for real-world document images.

document dewarping visual understanding robust perception

Statistics for AI

UltraTWD: Optimizing Ultrametric Trees for Tree-Wasserstein Distance

ICML, 2025

Fangchen Yu , Yanzhen Chen , Jiaxing Wei , Jianfeng Mao , Wenye Li^† , Qiang Sun^†

A method for optimizing ultrametric trees to improve efficient Tree-Wasserstein distance approximation.

tree-wasserstein distance ultrametric trees optimization

Statistics for AI

A Theory-Driven Approach to Inner Product Matrix Estimation for Incomplete Data: An Eigenvalue Perspective

WWW, 2025

Fangchen Yu^*† , Yicheng Zeng^*† , Jianfeng Mao , Wenye Li

A theory-driven method for inner product matrix estimation on incomplete data.

incomplete data inner product matrix eigenvalue estimation

Statistics for AI

Balancing the Trade-Off Between Global and Personalized Performance in Federated Learning

Information Sciences, 2025

Zibin Pan , Chi Li , Fangchen Yu , Shuyi Wang , Xiaoying Tang^† , Junhua Zhao^†

A federated learning method balances the trade-off between global and personalized performance.

federated learning personalization trade-off

Statistics for AI

KAE: Kolmogorov-Arnold Auto-Encoder for Representation Learning

arXiv, 2024

Fangchen Yu , Ruilizhen Hu , Yidong Lin , Yuqi Ma , Zhenghao Huang , Wenye Li^†

An auto-encoder architecture explores Kolmogorov-Arnold network for representation learning.

auto-encoder Kolmogorov-Arnold network representation learning

Statistics for AI

FedLF: Layer-wise Fair Federated Learning

AAAI, 2024

Zibin Pan , Chi Li , Fangchen Yu , Shuyi Wang ... Xiaoying Tang^† , Junhua Zhao^†

A layer-wise approach improves fairness in federated learning.

federated learning fairness layer-wise approach

Statistics for AI

Boosting Spectral Clustering on Incomplete Data via Kernel Correction and Affinity Learning

NeurIPS, 2023

Fangchen Yu , Runze Zhao , Zhan Shi , Yiwen Lu ... Jianfeng Mao , Wenye Li^†

Kernel correction and affinity learning improve spectral clustering with incomplete observations.

spectral clustering incomplete data kernel correction

Statistics for AI

Online Estimation of Similarity Matrices with Incomplete Data

UAI, 2023

Fangchen Yu , Yicheng Zeng , Jianfeng Mao , Wenye Li^†

A series of online estimation methods correct similarity matrices under incomplete data observations.

similarity matrix online estimation incomplete data

Statistics for AI

Highly-Efficient Robinson-Foulds Distance Estimation with Matrix Correction

ECAI, 2023

Fangchen Yu , Rui Bao , Jianfeng Mao , Wenye Li^†

Matrix correction enables efficient estimation of Robinson-Foulds distance.

Robinson-Foulds distance matrix correction tree distance

Statistics for AI

From Incompleteness to Unity: A Framework for Multi-view Clustering with Missing Values

ICONIP, 2023

Fangchen Yu , Zhan Shi , Yuqi Ma , Jianfeng Mao , Wenye Li^†

A multi-view clustering framework handles missing values across multiple views.

multi-view clustering missing values incomplete clustering

Statistics for AI

Metric Nearness Made Practical

AAAI, 2023

Wenye Li , Fangchen Yu , Zichen Ma

A practical approach improves the usability and efficiency of metric nearness model computation.

metric nearness model distance metric optimization

Statistics for AI

Calibrating Distance Metrics Under Uncertainty

ECML, 2022

Wenye Li^† , Fangchen Yu

A calibration method improves distance metric estimation when observations are uncertain.

distance metric uncertainty calibration method

Statistics for AI

Learning Sparse Binary Code for Maximum Inner Product Search

CIKM, 2021

Changyi Ma , Fangchen Yu , Yueyao Yu , Wenye Li^†

A sparse binary hashing method for maximum inner product search to preserve the pairwise similarities.

sparse binary code similarity search maximum inner product search