LongCat-DeepResearch Technical Report
LongCat-DeepResearch uses multi-agent planning and section-level revision to produce evidence-grounded research reports.
LongCat-DeepResearch combines an enhanced LongCat model with a multi-agent workflow that separates global planning from detailed investigation. Planning agents refine an actionable ResearchSpec, research agents draft assigned sections in parallel, and global review drives local revisions rather than full-report rewrites. It scores 55.25 on DeepResearchBench, 51.35 on DeepResearchBench II, and 79.83 on ResearchRubrics, plus 76.04 on an in-house benchmark, ranking second of four systems. Development analyses find benefits from combining planning perspectives, mixed effects from further plan refinement, and readability gains from additional editing.
- Planning agents build a ResearchSpec; research agents draft sections in parallel.
- Global review then applies targeted section-level revisions instead of full rewrites.
- Scores 55.25 on DeepResearchBench, 51.35 on DeepResearchBench II, and 79.83 on ResearchRubrics.
- In-house score is 76.04, second of four compared systems.
- Extra editing raises average automatic readability preference, with mixed trends by benchmark.
- AI models
- LongCatLongCat-DeepResearch
From the original article
1 min read at huggingface.co
We present LongCat-DeepResearch, a deep research system that combines an enhanced LongCat model with a multi-agent workflow for producing comprehensive, evidence-grounded reports. The workflow separates global planning from detailed investigation and coordinates revision at the section level. Multiple planning agents first explore external sources and refine an actionable research plan, termed ResearchSpec.
By Meituan LongCat Team, He Zhu, Yue Xu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.