人类研发时代结束？XYZ最强Search Agent横扫七大榜单，300个Agent跑通AI4AI全栈闭环
Ranking
Overall
71
Content
80
Popularity
N/A
No observed public metrics; popularity remains neutral/archived.
Merged summary
TL;DR - XYZ AI Lab released two Deep Search agents—35B-parameter Aquila-mini and 397B-parameter Aquila-pro—developed through an AI4AI workflow that coordinated over 300 agents. The release demonstrates bounded, human-governed AI participation across data generation, training, runtime optimization, and evaluation.
- Aquila-mini reportedly led same-scale systems across seven search benchmarks; Aquila-pro posted top public scores on BrowseComp-ZH, LiveBrowseComp, and WideSearch.
- Human-defined goals, budgets, permissions, and acceptance rules constrain agents, while independent evaluators and gates approve changes and escalate risks.
- The system retains append-only audit trails and shared memory so successful and failed experiments remain reproducible and reusable.
- Agent-discovered improvements included proactive context organization to preserve evidence during long search trajectories and state-faithful supervised fine-tuning based on information actually visible at each step.
Sources (1)
人类研发时代结束？XYZ最强Search Agent横扫七大榜单，300个Agent跑通AI4AI全栈闭环
Public signals
N/A
TL;DR - XYZ AI Lab released two Deep Search agents—35B-parameter Aquila-mini and 397B-parameter Aquila-pro—developed through an AI4AI workflow that coordinated over 300 agents. The release demonstrates bounded, human-governed AI participation across data generation, training, runtime optimization, and evaluation.
- Aquila-mini reportedly led same-scale systems across seven search benchmarks; Aquila-pro posted top public scores on BrowseComp-ZH, LiveBrowseComp, and WideSearch.
- Human-defined goals, budgets, permissions, and acceptance rules constrain agents, while independent evaluators and gates approve changes and escalate risks.
- The system retains append-only audit trails and shared memory so successful and failed experiments remain reproducible and reusable.
- Agent-discovered improvements included proactive context organization to preserve evidence during long search trajectories and state-faithful supervised fine-tuning based on information actually visible at each step.