Summary
Navalia is a framework for building LLM-based Cowork Agents that combines verifiable task synthesis with long-horizon post-training, targeting financial analysis tasks that require cross-document evidence integration and multistep numerical reasoning. Instead of costly expert task authoring or unreliable direct LLM synthesis, Navalia uses corporate financial reports as its data source: it semantically selects supporting evidence from complete fact sets, constructs executable computation chains, and generates analytical tasks with traceable evidence and real data dependencies between steps. Independent LLM reviews 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). Initialized from Qwen3.6-35B-A3B, Navalia achieves consistent improvements over the base model on five benchmarks: OfficeQA Pro, FinSearchComp, Finance Agent Benchmark, APEX-Agents-AA, and Workspace-Bench-Lite. Results indicate improved evidence integration, tool use, and long-horizon task execution, generalizing to broader cowork scenarios including general office tasks.
Navalia: Advancing Cowork Agents through Verifiable Task Synthesis and Long-Horizon Post-Training
Field: Machine Learning
Authors: Hengqi Liu, Ziyuan Liu, Yang Qin, Wanting Zhou, Zichuan Wang, Jiachen Liang, Xu Chu, Shaowei Chen, Zhaokai Luo, Mu Chuan
Published: 2026-10-22
arXiv: 2610.26887
Overview
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 requires these agents to integrate cross-document evidence and perform multistep numerical reasoning, but expert task authoring is costly, while direct LLM synthesis struggles to produce challenging tasks with reliable answers.
Navalia is a framework for Cowork Agents that combines verifiable task synthesis with long-horizon post-training:
- Data source: corporate financial reports. Navalia semantically selects supporting evidence from complete fact sets and constructs executable computation chains, generating analytical tasks with traceable evidence and actual data dependencies between steps.
- Verification: independent LLM review plus programmatic checks validate evidence consistency and answer correctness.
- Post-training: verified tasks are used for trajectory collection, cold-start supervised fine-tuning (SFT), and reinforcement learning (RL) to improve long-horizon task execution.
Results
Initialized with Qwen3.6-35B-A3B, Navalia achieves consistent improvements over the base model on five benchmarks:
- OfficeQA Pro
- FinSearchComp
- Finance Agent Benchmark
- APEX-Agents-AA
- Workspace-Bench-Lite
These results show that Navalia improves evidence integration, tool use, and long-horizon task execution, and generalizes to broader cowork scenarios including general office tasks.
---
*Auto-collected on 2026-10-22*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178636283