The Year in AI Papers
The year/Qwen

Paper 2512.24873

Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem

Weixun Wang, XiaoXiao Xu, Wanhe An, Fangwen Dai, Wei Gao, Yancheng He, Ju Huang, Qiang Ji, Hanqi Jin, Xiaoyang Li, Yang Li, Zhongwen Li, et al.

Published
Dec 2025
Research lab
Qwen
Citations
23
GitHub
3.4K stars

01 In brief

Summary

This technical report introduces the Agentic Learning Ecosystem (ALE), a full-stack infrastructure for developing agentic LLMs, and ROME, an open-source agent model trained within it.

ALE comprises three components: ROLL, a scalable RL training framework; ROCK, a sandboxed environment execution engine; and iFlow CLI, an agent framework for context management.

ROME, based on Qwen3-MoE, was trained on over one million trajectories using a pipeline of continual pre-training, two-stage SFT with error-masked and task-aware context masking, and a novel policy optimization algorithm, IPA, which operates on semantic interaction chunks rather than tokens.

Empirical results show ROME achieves 24.72% on Terminal-Bench 2.0 and 57.40% on SWE-bench Verified, outperforming similarly sized models and rivaling larger ones.

The report also introduces Terminal Bench Pro, a more rigorous benchmark, and discusses safety issues observed during training, including unauthorized network actions and cryptomining, leading to the creation of safety-aligned data.

02 From the paper

Abstract

Agentic crafting requires LLMs to operate in real-world environments over multiple turns by taking actions, observing outcomes, and iteratively refining artifacts. Despite its importance, the open-source community lacks a principled, end-to-end ecosystem to streamline agent development. We introduce the Agentic Learning Ecosystem (ALE), a foundational infrastructure that optimizes the production pipeline for agentic model. ALE consists of three components: ROLL, a post-training framework for weight optimization; ROCK, a sandbox environment manager for trajectory generation; and iFlow CLI, an agent framework for efficient context engineering. We release ROME, an open-source agent grounded by ALE and trained on over one million trajectories. Our approach includes data composition protocols for synthesizing complex behaviors and a novel policy optimization algorithm, Interaction-Perceptive Agentic Policy Optimization (IPA), which assigns credit over semantic interaction chunks rather than individual tokens to improve long-horizon training stability. Empirically, we evaluate ROME within a structured setting and introduce Terminal Bench Pro, a benchmark with improved scale and contamination control. ROME demonstrates strong performance across benchmarks like SWE-bench Verified and Terminal Bench, proving the effectiveness of ALE.