TBNAI
Navalia: Advancing Cowork Agents through Verifiable Task Synthesis and Long-Horizon Post-Training
Publication: cs.AI updates on arXiv.org
Author: Hengqi Liu, Ziyuan Liu, Yang Qin, Wanting Zhou, Zichuan Wang, Jiachen Liang, Xu Chu, Shaowei Chen, Zhaokai Luo, Mu Chuan
Source published: 21/10/2026, 04:00:00 UTC
Briefing created: 21/10/2026, 04:12:52 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 21/10/2026, 04:03:24 UTC.
Briefings
Navalia is a framework for Cowork Agents that combines verifiable task synthesis with long-horizon post-training. Using corporate financial reports as data, it semantically selects evidence and constructs executable computation chains to generate analytical tasks with traceable dependencies. The system uses independent LLM review and programmatic checks to verify evidence consistency and answer correctness before training. Navalia, initialized from Qwen3.6-35B-A3B, achieved consistent gains across five benchmarks including OfficeQA Pro, FinSearchComp, and Finance Agent Benchmark, demonstrating improved evidence integration and long-horizon task execution.
Key points
- Navalia uses corporate financial reports as its primary data source to semantically select evidence and construct executable computation chains.
- The framework achieves consistent gains across five benchmarks: OfficeQA Pro, FinSearchComp, Finance Agent Benchmark, APEX-Agents-AA, and Workspace-Bench-Lite.
Context
Navalia addresses the need for autonomous agents to integrate cross-document evidence and perform multistep numerical reasoning in financial analysis, where expert task authoring is costly and direct LLM synthesis struggles with reliable task generation.
Limitations
Evidence is limited to the single preprint arXiv:2610.26887v1 from October 2026. Results are based on initial training and benchmarks; long-horizon performance and generalization to broader office tasks beyond financial reports remain unverified.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 21/10/2026, 04:12:52 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeieu63z5orudgmthe4al6snhd7klg7qrdrqr7dpt7m5dkfv27amaue