Meituan Launches Native Deep Research Agent
Recently, Meituan's LongCat platform officially launched a native 'Deep Research' agent. This new AI model, built on a proprietary three-layer closed-loop architecture and trained on real-world data, is designed to tackle the common issue of AI hallucinations in practical scenarios.
Three-Layer System Enhances Accuracy
The new agent features a three-tiered architecture, including perception, reasoning, and feedback layers. This system allows AI to continuously verify data sources during complex task processing, significantly reducing the risk of inaccurate or hallucinated outputs.
- Perception Layer: Captures real-world data
- Reasoning Layer: Analyzes current and historical data
- Feedback Layer: Continuously improves outputs
Blind Testing Shows Superior Performance
In a recent round of blind user testing, the performance of Meituan's Deep Research Agent was impressive. Its research output had an overall usability rate of 61.1%, significantly outperforming ChatGPT's 42.8%. This demonstrates its strong practical value and precision.
Available for Free
The agent is now freely accessible via the LongCat web platform. Users can experience its powerful research capabilities without registration. It is expected to be widely applied in travel, consumption, and local life services in the near future.