ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents

Hugging Face Daily Papers Papers
coding-agents training-environments program-synthesis long-horizon-agents reinforcement-training fine-tuning

Summary

ProgramGym introduces a framework with Program-Env and Program-Agent for building scalable, verifiable training environments for long-horizon program reproduction agents, achieving notable performance gains when fine-tuning Qwen3.8-27B on the generated environments.

Coding agents are increasingly used to build programs from scratch as software engineers do rather than to resolve issues within a repository or to generate a repository from a given requirements document. However, scalable verifiable training environments that force agents to discover what is required and to decide how the system should be designed remain scarce. Moreover, existing agent workflows also cannot sustain long-horizon development in which the agent explores the reference program before implementation. To address these challenges, we introduce ProgramGym, a framework consisting of Program-Env and Program-Agent for scalable training environments of long-horizon program reproduction agents. Specifically, Program-Env constructs verifiable training environments from repositories through a feedback-driven loop that repairs runtime sandbox builds and the quality of black-box tests. Using this novel pipeline, we obtain 1,648 training environments spanning C, C++, Go, and Rust. Program-Agent combines a role-isolated workflow of analysis, implementation, and review with revision management for compilation checks and failure recovery. Experiments on ProgramBench show that both Program-Env and Program-Agent improve performance on program reproduction. Fine-tuning on trajectories with these constructed environments raises the performance of Qwen3.8-27B from 34.87% to 42.84%. Program-Agent further raises the base and SFT models to 55.71% and 56.25%, both exceeding some flagship models. We will release the ProgramGym to support further research on coding agents that build programs from scratch.
Original Article
View Cached Full Text

Cached at: 10/12/26, 02:38 AM

Paper page - ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents

Source: https://huggingface.co/papers/2610.13395 Authors:

,

,

,

,

,

,

,

,

,

,

,

Abstract

Codingagentsareincreasinglyusedtobuildprogramsfromscratchassoftwareengineersdoratherthantoresolveissueswithinarepositoryortogeneratearepositoryfromagivenrequirementsdocument.However,scalableverifiabletrainingenvironmentsthatforceagentstodiscoverwhatisrequiredandtodecidehowthesystemshouldbedesignedremainscarce.Moreover,existingagentworkflowsalsocannotsustainlong-horizondevelopmentinwhichtheagentexploresthereferenceprogrambeforeimplementation.Toaddressthesechallenges,weintroduceProgramGym,aframeworkconsistingofProgram-EnvandProgram-Agentforscalabletrainingenvironmentsoflong-horizonprogramreproductionagents.Specifically,Program-Envconstructsverifiabletrainingenvironmentsfromrepositoriesthroughafeedback-drivenloopthatrepairsruntimesandboxbuildsandthequalityofblack-boxtests.Usingthisnovelpipeline,weobtain1,648trainingenvironmentsspanningC,C++,Go,andRust.Program-Agentcombinesarole-isolatedworkflowofanalysis,implementation,andreviewwithrevisionmanagementforcompilationchecksandfailurerecovery.ExperimentsonProgramBenchshowthatbothProgram-EnvandProgram-Agentimproveperformanceonprogramreproduction.Fine-tuningontrajectorieswiththeseconstructedenvironmentsraisestheperformanceofQwen3.8-27Bfrom34.87%to42.84%.Program-AgentfurtherraisesthebaseandSFTmodelsto55.71%and56.25%,bothexceedingsomeflagshipmodels.WewillreleasetheProgramGymtosupportfurtherresearchoncodingagentsthatbuildprogramsfromscratch.

View arXiv pageView PDFProject pageAdd to collection

Get this paper in your agent:

hf papers read 2610\.13395

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.13395 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.13395 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.13395 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

SkillGym: Training Skill-Use Agents with Automatic Verifiable Environment Generation

Hugging Face Daily Papers

SkillGym proposes an automatic pipeline that crawls reproducible skills from the internet, builds verifiable difficulty-controlled environments, and collects 19k verified trajectories to train skill-use agents. Fine-tuning Qwen3.5 models (2B to 122B) on these trajectories improves performance across four skill-use benchmarks, with the 9B SFT model outperforming a 397B untrained model on two of them.

Cowork-Gym: Synthesizing Interactive Work Environments at Scale

Hugging Face Daily Papers

Cowork-Gym 通过 build-execute-revise 循环与有状态用户模拟器大规模合成协作式 agent 环境（64.9K 个、14 个应用场景），并用于后训练 9B/27B/35B-A3B Qwen 模型，在六个 agentic 基准上分别提升 8.7、6.4、4.1 分，其中 27B 版本在 PawBench 和 ClawEval 上取得最高平均分。

AutoGym: Blueprint-First Generation of Verifiable Agent Gyms

arXiv cs.AI

AutoGym presents a framework for automatically generating complete agent gyms—tasks, executable environments, and verifiers—from minimal seeds using blueprint-first generation to ensure solvability and active curriculum synthesis for adaptive difficulty steering.

AutoEnvScaling: Automating the Data Flywheel with Terminal Agents

Hugging Face Daily Papers

AutoEnvScaling treats RL environment design as a terminal task, letting a proposer agent automatically build training environments from solver rollouts to automate the data flywheel. The method raises the valid-environment rate up to 25x at 75% lower cost, improves Qwen terminal-agent scores by up to 12.2 points, and enables a single model to co-evolve as both solver and generator, a step toward recursive self-improvement.