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

2026-02-03

AI Summary

This work addresses the scarcity of high-quality, verifiable training data and the high cost of manual annotation in complex reasoning tasks by proposing the Agentic Proposing framework, which formulates problem synthesis as a goal-driven sequential decision-making process. By dynamically composing modular reasoning skills, integrating internal reflection with tool invocation, the framework generates precise and verifiable training trajectories. It further introduces Multi-Granularity Policy Optimization (MGPO) to enhance synthesis efficiency. A 30B-parameter model trained solely on 11,000 such synthetic trajectories achieves 91.6% accuracy on AIME25, matching the performance of GPT-5 and significantly outperforming existing baselines. These results demonstrate that small-scale, high-quality synthetic data can effectively replace large-scale human-annotated datasets while exhibiting strong cross-domain generalization capabilities.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Multiagent Systems: Multiagent Learning
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

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.

Problem

• complex reasoning
• synthetic data
• large language models
• data scalability
• verifiable datasets

Innovation

• Agentic Proposing
• Compositional Skill Synthesis
• Multi-Granularity Policy Optimization
• Synthetic Data Generation
• Verifiable Reasoning Trajectories