---
format: "aidr-story-markdown/v1"
id: "4a2882d30389b2a6b571b1f4b718c6130d7d23cb442c0edfa8df6536c8ccc081"
canonical_url: "https://aidr.today/4a2882d3?lang=en"
title: "ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-12T00:00:00.000Z"
category: "Frameworks"
topics: ["agent","coding","framework","llm","benchmark"]
source_urls: ["https://huggingface.co/papers/2610.13395"]
summary: "ProgramGym is a framework consisting of Program-Env and Program-Agent designed for training long-horizon program reproduction agents. Program-Env creates 1,648 verifiable training environments across C, C++, Go, and Rust using a feedback-driven loop. Program-Agent utilizes a role-isolated workflow of analysis, implementation, and review with revision management."
---

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

> [Open the canonical story](<https://aidr.today/4a2882d3?lang=en>)

**Published:** 2026-10-12T00:00:00.000Z
**Category:** Frameworks
**Topics:** agent, coding, framework, llm, benchmark

## Summary

ProgramGym is a framework consisting of Program\-Env and Program\-Agent designed for training long\-horizon program reproduction agents\. Program\-Env creates 1,648 verifiable training environments across C, C\+\+, Go, and Rust using a feedback\-driven loop\. Program\-Agent utilizes a role\-isolated workflow of analysis, implementation, and review with revision management\.

## Sources

- [Story source](<https://huggingface.co/papers/2610.13395>)