国外办了场AI投资实盘大赛，国产大模型目前断档式领先

Core Insights - The article discusses a project called "Alpha Arena" initiated by a foreign AI laboratory named nof1, which pits six advanced AI models against each other in real-time trading with a starting capital of $10,000 each, aiming to test their investment strategies and performance in the financial market [2][33]. Group 1: Performance of AI Models - As of October 25, Qwen3 MAX leads with a 49% return, followed by DeepSeek at 13%, while other models like Gemini 2.5 Pro and GPT-5 show significant losses of -67% and -75% respectively [3][4][6]. - The trading competition has seen dramatic fluctuations, with DeepSeek initially leading but later overtaken by Qwen3 MAX, showcasing the volatility and unpredictability of AI-driven trading [12][29]. - The performance of the models varies significantly, with DeepSeek adopting a long-term investment strategy similar to value investing, while Gemini 2.5 Pro exhibits a high-frequency trading approach with an average holding time of only 2 hours and 29 minutes [20][17]. Group 2: Investment Strategies - DeepSeek employs a straightforward investment strategy, focusing on major cryptocurrencies like BTC and ETH, and maintains a median holding period of 38 hours and 32 minutes, indicating a more stable approach [18][17]. - In contrast, Gemini 2.5 Pro's strategy is erratic, characterized by frequent trades and a lack of consistent direction, leading to poor performance [20]. - Qwen3 MAX adopts an aggressive strategy, often going "all in" on a single asset with high leverage, resulting in high volatility and potential for significant gains or losses [27][28]. Group 3: Implications for AI in Finance - The competition serves as a "financial Turing test," aiming to determine whether AI can outperform human financial experts in a complex and uncertain environment [33][34]. - The rise of AI-driven trading is highlighted, with statistics showing that a significant portion of trading volume in cryptocurrency and stock markets is already automated, indicating a shift towards algorithmic trading [35][36]. - The article raises concerns about the potential risks of widespread adoption of similar AI models, suggesting that if many traders use the same strategies, it could lead to market instability during adverse conditions [40][41].