Latest evaluation data on seven major models for paper trading U.S. stocks has been released, yielding highly counterintuitive results: The largest general-purpose model, while quick to catch market trends, suffers severely from overtrading in volatile markets—recommending portfolio adjustments at the slightest news headline, with drawdowns peaking above 12%. In contrast, the finance-tuned model and the hybrid decision tree (XGBoost + RAG) achieved a Sharpe ratio of 1.45, keeping maximum drawdown firmly under 8%. The conclusion is clear: Don’t rely solely on a single large model to generate direct buy/sell signals for U.S. stocks. Use Claude to review lengthy earnings reports, leave signal filtering to mathematical models—combining these approaches is the right strategy. https://t.co/WUz7XM8ERC (GoBull 2026: Real-World Benchmark Test Report on Seven Major Models for U.S. Stock Quantitative Trading) (⚠️ Disclaimer: The above is for academic testing and strategy discussion only and does not constitute investment advice for U.S. stocks.) #USStocks #AIInvesting