Saved in:
Bibliographic Details
Main Authors: He, Runyuan, Mang, Qiuyang, Zhou, Shang, Liu, Kaiyuan, Li, Hanchen, Mao, Huanzhi, Zhang, Qizheng, Li, Zerui, Peng, Bo, Cheng, Lufeng, Fu, Tianfu, Wang, Yichuan, Chai, Wenhao, Shang, Jingbo, Dimakis, Alex, Gonzalez, Joseph E., Cheung, Alvin
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.14445
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918501154816000
author He, Runyuan
Mang, Qiuyang
Zhou, Shang
Liu, Kaiyuan
Li, Hanchen
Mao, Huanzhi
Zhang, Qizheng
Li, Zerui
Peng, Bo
Cheng, Lufeng
Fu, Tianfu
Wang, Yichuan
Chai, Wenhao
Shang, Jingbo
Dimakis, Alex
Gonzalez, Joseph E.
Cheung, Alvin
author_facet He, Runyuan
Mang, Qiuyang
Zhou, Shang
Liu, Kaiyuan
Li, Hanchen
Mao, Huanzhi
Zhang, Qizheng
Li, Zerui
Peng, Bo
Cheng, Lufeng
Fu, Tianfu
Wang, Yichuan
Chai, Wenhao
Shang, Jingbo
Dimakis, Alex
Gonzalez, Joseph E.
Cheung, Alvin
contents Many real-world coding challenges are open-ended and admit no known optimal solution. Yet, recent progress in LLM coding has focused on well-defined tasks such as feature implementation, bug fixing, and competitive programming. Open-ended coding remains a weak spot for LLMs, largely because open-ended training problems are scarce and expensive to construct. Our goal is to synthesize open-ended coding problems at scale to train stronger LLM coders. We introduce FrontierSmith, an automated system for iteratively evolving open-ended problems from existing closed-ended coding tasks. Starting from competitive programming problems, FrontierSmith generates candidate open-ended variants by changing the problems'goals, restricting outputs, and generalizing inputs. It then uses a quantitative idea divergence metric to select problems that elicit genuinely diverse approaches from different solvers. Agents then generate test cases and verifiers for the surviving candidates. On two open-ended coding benchmarks, training on our synthesized data yields substantial gains over the base models: Qwen3.5-9B improves by +8.82 score on FrontierCS and +306.36 (Elo-rating-based performance) on ALE-bench; Qwen3.5-27B improves by +12.12 and +309.12, respectively. The synthesized problems also make agents take more turns and use more tokens, similar to human-curated ones, suggesting that closed-ended seeds can be a practical starting point for long-horizon coding data.
format Preprint
id arxiv_https___arxiv_org_abs_2605_14445
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle FrontierSmith: Synthesizing Open-Ended Coding Problems at Scale
He, Runyuan
Mang, Qiuyang
Zhou, Shang
Liu, Kaiyuan
Li, Hanchen
Mao, Huanzhi
Zhang, Qizheng
Li, Zerui
Peng, Bo
Cheng, Lufeng
Fu, Tianfu
Wang, Yichuan
Chai, Wenhao
Shang, Jingbo
Dimakis, Alex
Gonzalez, Joseph E.
Cheung, Alvin
Machine Learning
Many real-world coding challenges are open-ended and admit no known optimal solution. Yet, recent progress in LLM coding has focused on well-defined tasks such as feature implementation, bug fixing, and competitive programming. Open-ended coding remains a weak spot for LLMs, largely because open-ended training problems are scarce and expensive to construct. Our goal is to synthesize open-ended coding problems at scale to train stronger LLM coders. We introduce FrontierSmith, an automated system for iteratively evolving open-ended problems from existing closed-ended coding tasks. Starting from competitive programming problems, FrontierSmith generates candidate open-ended variants by changing the problems'goals, restricting outputs, and generalizing inputs. It then uses a quantitative idea divergence metric to select problems that elicit genuinely diverse approaches from different solvers. Agents then generate test cases and verifiers for the surviving candidates. On two open-ended coding benchmarks, training on our synthesized data yields substantial gains over the base models: Qwen3.5-9B improves by +8.82 score on FrontierCS and +306.36 (Elo-rating-based performance) on ALE-bench; Qwen3.5-27B improves by +12.12 and +309.12, respectively. The synthesized problems also make agents take more turns and use more tokens, similar to human-curated ones, suggesting that closed-ended seeds can be a practical starting point for long-horizon coding data.
title FrontierSmith: Synthesizing Open-Ended Coding Problems at Scale
topic Machine Learning
url https://arxiv.org/abs/2605.14445