AgenticMath¶
Product or Programme · All products and programmes
Evidence · 4 claims (4 for) · 1 study (1 design),q2· 0 of 1 report an effect size · 4 claims rest on one study
Description¶
A Liu et al. automated multi-agent framework that filters seed problems, iteratively generates and reviews new problems, produces solutions, and ranks synthetic question-answer data for LLM supervised fine-tuning.
Components¶
- AgenticMath four-stage multi-agent data generation framework: AgenticMath is an automated multi-agent framework enforcing quality control at every stage of mathematical data generation. It proceeds in four stages: seed problem filtering (LLM scoring on complexity, information value, and clarity with k-NN score curation), agentic problem generation (rephrase, review, and revise agents with an iterative loop), solution generation (one-shot chain-of-thought prompting), and synthetic-data evaluation with ranking-based selection. The article describes it as "an automated multi-agent framework that enforces quality control at every stage of mathematical data generation," following a "Less Is More" principle to yield a data-efficient, high-quality dataset. (Xianyang Liu et al. (2025))
Claims¶
- Each AgenticMath component contributes performance gains, with the review-revise loop playing a particularly important role [+M]
- Higher seed-problem filtering thresholds yield better downstream reasoning performance, indicating problem quality directly impacts results [+M]
- Increasing AgenticMath-generated training data from 7.5K to 22.5K samples yields consistent accuracy gains for Llama3-8B across benchmarks [+W]
- At 30K and 60K data scales, AgenticMath outperforms all compared baselines across four base models on six math benchmarks [+W]
Related Products and Programmes¶
-
Key Sources¶
- Xianyang Liu, Yilin Liu, Shuai Wang, Hao Cheng, Andrew Estornell, Yuzhi Zhao, Jun Shu, Jiaheng Wei. (2025). AgenticMath: Enhancing LLM Reasoning via Agentic-based Math Data Generation. arXiv preprint. https://arxiv.org/abs/2510.19361
Page metadata
| Field | Value |
|---|---|
type |
product |
id |
agenticmath |
title |
AgenticMath |
description |
A Liu et al. |
status |
review |
product_kind |
software |
generated |
by process:settle-candidates on 2026-10-11 |
trust |
unverified — no one has confirmed this page's content |
Frontmatter is the page's machine-readable record — see the Schema & Guide.