Agentic Proposing: Enhancing Large Language Model Reasoning via Compositional Skill Synthesis

TL;DR

This paper introduces Agentic Proposing, a novel framework for synthesizing high-quality, verifiable training data for large language models by dynamically composing modular reasoning skills, leading to significant improvements in reasoning tasks.

Contribution

It presents a goal-driven, agent-based approach to problem synthesis that balances complexity and validity, enabling the creation of effective training data with fewer instances.

Findings

Models trained on synthesized data outperform baselines.

Achieves 91.6% accuracy on AIME25 with only 11,000 trajectories.

Synthetic data rivals proprietary models like GPT-5.

Abstract

Advancing complex reasoning in large language models relies on high-quality, verifiable datasets, yet human annotation remains cost-prohibitive and difficult to scale. Current synthesis paradigms often face a recurring trade-off: maintaining structural validity typically restricts problem complexity, while relaxing constraints to increase difficulty frequently leads to inconsistent or unsolvable instances. To address this, we propose Agentic Proposing, a framework that models problem synthesis as a goal-driven sequential decision process where a specialized agent dynamically selects and composes modular reasoning skills. Through an iterative workflow of internal reflection and tool-use, we develop the Agentic-Proposer-4B using Multi-Granularity Policy Optimization (MGPO) to generate high-precision, verifiable training trajectories across mathematics, coding, and science. Empirical results demonstrate that downstream solvers trained on agent-synthesized data significantly outperform leading baselines and exhibit robust cross-domain generalization. Notably, a 30B solver trained on only 11,000 synthesized trajectories achieves a state-of-the-art 91.6% accuracy on AIME25, rivaling frontier-scale proprietary models such as GPT-5 and proving that a small volume of high-quality synthetic signals can effectively substitute for massive human-curated datasets.