Navalia: Advancing Cowork Agents through Verifiable Task Synthesis and Long-Horizon Post-Training

Authors: Hengqi Liu, Ziyuan Liu, Yang Qin, Wanting Zhou, Zichuan Wang, Jiachen Liang, Xu Chu, Shaowei Chen, Zhaokai Luo, Mu Chuan

Subjects: Artificial Intelligence (cs.AI) | Submitted: 13 Oct 2026 | arXiv: 2610.26887 [cs.AI] | Comments: 13 pages, 2 figures

Abstract

Large language model (LLM) agents are evolving from conversational assistants into Cowork Agents capable of autonomous planning, tool use, and complex task execution. Financial analysis in particular requires these agents to integrate cross-document evidence and perform multistep numerical reasoning — yet expert task authoring is costly, while direct LLM synthesis struggles to produce challenging tasks with reliable answers.

This paper introduces Navalia, a framework for Cowork Agents that combines verifiable task synthesis with long-horizon post-training. Using corporate financial reports as its data source, Navalia semantically selects supporting evidence from complete fact sets and constructs executable computation chains to generate analytical tasks with traceable evidence and genuine data dependencies between steps. Independent LLM review and programmatic checks then verify evidence consistency and answer correctness.

The verified tasks are used for trajectory collection, cold-start supervised fine-tuning (SFT), and reinforcement learning (RL) to improve long-horizon task execution. Initialized from Qwen3.6-35B-A3B, Navalia achieves consistent gains over the base model across all five benchmarks: OfficeQA Pro, FinSearchComp, Finance Agent Benchmark, APEX-Agents-AA, and Workspace-Bench-Lite. These results demonstrate that Navalia improves evidence integration, tool use, and long-horizon task execution, and generalizes to broader Cowork settings, including general office tasks.

Key Contributions

• Verifiable task synthesis pipeline that grounds tasks in corporate financial reports, with traceable evidence and stepwise data dependencies.
• Dual verification via independent LLM review and programmatic checks to ensure evidence consistency and answer correctness.
• Long-horizon post-training recipe combining trajectory collection, cold-start SFT, and RL.
• Consistent empirical gains across five benchmarks, with demonstrated generalization to general office-task settings.

Citation

arXiv:2610.26887 [cs.AI]
https://arxiv.org/abs/2610.26887

via ArXiv AI