UCLA's PANINI-DR hits 73.3% on BrowseComp-Plus, beating best deep-research harness by 16 points
M. Yigit Turali · hf · 2026-10-14
UCLA researchers introduce PANINI-DR, a multi-agent deep research harness that separates investigation coordination from per-subquestion research and verification via a shared Generative Semantic Workspace recording subquestions, dependencies, findings, and evidence. Across BrowseComp-Plus, FRAMES, and MoNaCo, it beats seven state-of-the-art harnesses sharing a frozen model backbone: 73.3% accuracy on BrowseComp-Plus (+16.0 over next best), nearly 2x on hard questions, and the highest evidence recoverability from saved citations. Ablations confirm structured planning, persistent state, and per-finding evidence checks drive the gains.
More from coding & agent
- exe.dev's first meetup in SF tackles 'Are We Running Too Many Agents?' — zeeg · 2026-10-14
- When Agents Outrun the Org Chart: Mik Kersten's SAFe Summit Keynote — RealGeneKim · 2026-10-14
- Google 提出 Harness²：递归自进化 agent 执行框架，最高涨 11.8 分 — google · 2026-10-14
- Agent Sociodrama: fixing LLM agent teams via social dynamics, up to 83.3pp accuracy gain — nanyang-technological-university-singapore · 2026-10-14
- The Structure Agents Need Isn't Technical — It's Operational — usehive · 2026-10-14
- Claude Code 2.1.299 ships 190 CLI changes, adds idle compaction and --disallow-tools — ClaudeCodeLog · 2026-10-14