Recent tests of AI trading models in real market conditions have revealed mixed performance, highlighting both potential and limitations. The AI-Trader framework was used to evaluate the financial decision-making capabilities of leading language models across different asset pools, including Nasdaq-100 components, SSE 50 components, and top cryptocurrency assets. The competition, held from November 25 to November 7, saw MiniMax-M2 excel in U.S. stocks and A-shares, while DS-V3.1 led in cryptocurrencies. Despite some successes, most AI models struggled with low returns and weak risk management in real markets. The performance varied significantly across different markets, with models like MiniMax-M2 adapting strategies based on market conditions. However, challenges such as misanalysis, frequent trading, and inadequate risk control were prevalent, underscoring the need for improved information verification and error correction in AI trading systems.