Lemon Agent Technical Report
A new multi-agent AI system sets records on major benchmarks by intelligently managing its resources.
Deep Dive
Researchers have introduced Lemon Agent, a new AI system designed to manage teams of specialized AI agents. It uses a two-tier 'orchestrator-worker' framework to dynamically allocate computational power and run subtasks in parallel, improving efficiency. The system also features advanced memory and context management. It achieved a state-of-the-art 91.36% accuracy on the GAIA benchmark and a top score of 77+ on xbench-DeepSearch, demonstrating superior performance on complex, multi-step problems.
Why It Matters
This approach makes advanced AI more practical and cost-effective for real-world, long-term tasks like research and analysis.