A newly published technical report presents LongCat-DeepResearch, a deep research system that pairs an enhanced LongCat model with a multi-agent workflow designed to produce comprehensive, evidence-grounded reports.
The workflow divides labor between global planning and detailed investigation, and coordinates revision at the section level. Multiple planning agents first explore external sources and refine an actionable research plan the authors call a ResearchSpec. Research agents then investigate and draft their assigned sections in parallel, gathering additional evidence in separate contexts as their analyses develop.
After the sections are assembled, a global review step guides targeted local revisions, which the report says reduces reliance on repeatedly rewriting the entire report. The same workflow also supports building research tasks and trajectories for mid-training and post-training of LongCat's general-purpose models.
On reported benchmarks, LongCat-DeepResearch scores 55.25 on DeepResearchBench, 51.35 on DeepResearchBench II, and 79.83 on ResearchRubrics. On an in-house benchmark it scores 76.04, ranking second among four compared systems.
Development-set analyses indicate benefits from combining planning perspectives, while further planning refinement shows mixed effects. Additional editing improves average automatic readability preference across two benchmarks, though the report notes differing trends on each.