AgentCPM-Report: Interleaving Drafting and Deepening for Open-Ended Deep Research

2026

Read AgentCPM-Report: Interleaving Drafting and Deepening for Open-Ended Deep Research (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-06540.txt.

Description

Deep research reports demand gathering large amounts of information and weaving it into analysis, yet most systems plan an outline first and then write to it. Yishan Li and 14 co-authors argue that outline-first writing is fragile: a weak outline caps the whole report, and the approach usually needs large closed models.

Their framework, AgentCPM-Report, instead writes the way people do. A policy called Writing As Reasoning lets the model revise its outline as it drafts and digs deeper, alternating evidence-based drafting with reasoning-driven deepening. A multi-stage training recipe teaches a compact eight-billion-parameter agent that skill. The local system aims to lower deployment barriers and keep user data private.

Catalog record

YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE58 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-06540

Browse the full library · About the Tower of Babel