How Does LLM Trade Agents Work In The Real World? Six Month Production Data Announced
Trade systems managed by artificial intelligence are now going beyond laboratory experiments and trading in the real world. A new research published in arXiv offers comprehensive data obtained from six-month production experience of trade agents controlled by large language models (LLM).
Research covered two different platforms and user base:
- DX Terminal Pro: Vaults funded by 3,505 users traded with real ETH in the base blockchain on February-Mart 2026
- DXAP live alpha fleet: The agents created by 500-599 users were processed in Hyperliquid perpetual futures markets in June-August 2026
- In simultaneous active 91-117 agent, total 500-599 agents were used throughout the entire period
- Continuous measurement record significantly reveals the behavior of agents in real conditions
This study provides the first systematic data on**, how **LLM-based trading systems perform on the scale. It is critical to understand how these technologies often discussed in the academic environment have decided over millions of dollars of real transactions.
Research results contain important findings about the reliability, effectiveness and risks of financial applications of AI agents. While the role of autonomous systems in the blockchain and DeFi ecosystem is growing, such real world data will become the basic reference in the decisions of industry and researchers.