TagsSciAgent: A Unified Multi-Agent System for Generalistic Scientific Reasoning (Nov 2025)
- Link: http://arxiv.org/abs/2511.08151v1
- Date: November 2025
Abstract
SciAgent is a unified, hierarchical multi-agent AI system designed for generalistic scientific reasoning, enabling adaptability across diverse disciplines and difficulty levels. It features a Coordinator Agent for global task understanding and routing, specialized Worker Systems for domain-specific reasoning, and Sub-agents for execution. The system dynamically assembles and refines reasoning pipelines through feedback loops, achieving gold-medal-level performance in Mathematics and Physics Olympiads (IMO, IMC, IPhO, CPhO) and demonstrating strong generalization on IChO and Humanity’s Last Exam benchmarks. This work establishes a concrete step towards AI systems capable of coherent, cross-disciplinary reasoning at expert levels.
Key Topics:
- Multi-agent systems
- Generalistic scientific reasoning
- Hierarchical AI architecture
- Olympiad problem solving
- Large Language Models (LLMs)
- Adaptive reasoning
- Domain generalization
- Scientific problem solving