2026 CAAI-Ant Group Research Fund (AGI Special Program)
2026年度CAAI-蚂蚁科研基金（AGI专项）项目
June 9, 2026
Sponsor:
Chinese Association for Artificial Intelligence
The official announcements in Chinese and English.
About:
The "CAAI-Ant Group Research Fund (AGI Special Program)" is jointly established by CAAI and Ant Group to support industry-academia collaboration and the co-development of the AGI field. It aims to build a platform for industry-academia cooperation, connect industrial practice with academic research, and support outstanding scholars worldwide in conducting cutting-edge research that integrates with industry needs.
Funding Scale and Period:
Selected projects will receive funding of RMB 300,000 per project. Each project has a one-year cooperation period, with the possibility of rolling support for projects that achieve outstanding collaboration outcomes.
Funding Area:
The application covers three major research directions: AI Foundation Models, Agentic AI / AI Agents, and Token Efficiency Optimization / World Models. A total of 26 research topics are available.
AI Foundation Models
- High-Precision Numerical Understanding and Analysis for Finance
- Unified Representation and Cross-Modal Alignment for Generalist Multimodal Foundation Models
- ModelGrow: Continuous Evolution of Large Models Through Online Feedback Loops
- Unified Cross-Modal Representation for Voice-to-Singing Timbre Conversion in Low-Latency Streaming Scenarios
- Unified Multimodal Understanding and Generation Models for Visual Guidance
- OCR Technologies Based on Multimodal Large Models
- Evaluation-Driven Fine-Grained Controllable UI Infographic Generation and Editing
- Structured Image–Text Corpus Synthesis
- Mental Health Intervention Systems Based on Long-Term Memory and Dynamic Tool Use in Large Models
Agentic AI/AI Agent
- Adaptive Retrieval Planning and Multi-Round Knowledge Acquisition for Complex-Task Agents
- A Cooperative RL–OPD Training System for Agentic AI
- A Multi-Dimensional Controllability Evaluation Framework for Reliable and Controllable AI Agents
- An Agentic RL Decision-Optimization Framework for Advertising: From Intelligent Bidding to End-to-End Autonomous Delivery
- Agentic Learning for Long-Horizon Tasks: Development and Deployment
- Construction and Training of General-Purpose Game Agents Based on Multimodal Large Models
- Multimodal Agents for Complex Tasks
- Agentic RL Algorithms for Scientific Intelligence
- Reasoning Consistency and Hallucination Suppression in Deep Research Systems
- Evaluation and Optimization of LLM Adaptation to Agent Harnesses
- Training Methods for Reasoning and Planning in Code-Agent Tasks with Large Language Models
- Optimizing Long-Horizon Planning and Coding Capabilities for On-Device Code Agents
- Self-Evolving Learning and Optimization Methods for Intelligent Agents
- Key Technologies for Multi-Agent Collaboration
Token Efficiency Optimization/World Models
- Efficient Token Utilization in Large Model Pretraining: Architectures and Optimization Methods
- RDMA Multiplexing–Based Storage Traffic Acceleration
- World Model–Based Intelligent Interaction and Control for Desktop Robots
Eligibility:
- Applicant should be a full-time researcher (excluding post-docs) in higher education universities or research institutions.
- Applicant should have doctoral degree or be qualified to supervise graduate students.
- Applicant should be able to conduct research independently and lead groups
- Applicant needs to be CAAI members.
Applications:
Proposals should be submitted through https://www.antresearch.com/Cooperation/Funds/FundsDetails/20260522001 by June 30, 2026. If you are interested to apply, please also let us know at shanghai.researchgrants@nyu.edu.