ReThinker: Scientific Reasoning by Rethinking with Guided Reflection and Confidence Control

2026-02-04

AI Summary

This work addresses the limitations of large language models in expert-level scientific reasoning tasks—such as those in the HLE benchmark—where fixed tool pipelines, fragile multi-agent collaboration, and inefficient test-time scaling hinder performance. The authors propose ReThinker, a novel framework featuring a Solver-Critic-Selector architecture that enables confidence-driven dynamic computation allocation. ReThinker supports adaptive tool invocation, multi-dimensional guided reflection, and confidence-weighted selection. Key innovations include a confidence-aware reasoning mechanism, annotation-free reverse data synthesis, and an adaptive trajectory reuse strategy, collectively overcoming the constraints of conventional pipeline designs. Evaluated on HLE, GAIA, and XBench, ReThinker substantially outperforms both state-of-the-art base models and advanced research systems, establishing a new frontier in expert-level scientific reasoning.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Expert-level scientific reasoning remains challenging for large language models, particularly on benchmarks such as Humanity's Last Exam (HLE), where rigid tool pipelines, brittle multi-agent coordination, and inefficient test-time scaling often limit performance. We introduce ReThinker, a confidence-aware agentic framework that orchestrates retrieval, tool use, and multi-agent reasoning through a stage-wise Solver-Critic-Selector architecture. Rather than following a fixed pipeline, ReThinker dynamically allocates computation based on model confidence, enabling adaptive tool invocation, guided multi-dimensional reflection, and robust confidence-weighted selection. To support scalable training without human annotation, we further propose a reverse data synthesis pipeline and an adaptive trajectory recycling strategy that transform successful reasoning traces into high-quality supervision. Experiments on HLE, GAIA, and XBench demonstrate that ReThinker consistently outperforms state-of-the-art foundation models with tools and existing deep research systems, achieving state-of-the-art results on expert-level reasoning tasks.

Problem

• scientific reasoning
• large language models
• expert-level reasoning
• tool use
• multi-agent coordination

Innovation

• confidence-aware reasoning
• guided reflection
• adaptive tool use
• trajectory recycling
• solver-critic-selector architecture