Core Insights - The AI investment competition Alpha Arena concluded with Alibaba's Qwen achieving a remarkable return of over 20%, securing the championship title [1][21] - DeepSeek ranked second, marking the only two profitable models in the competition, while the four major US models suffered significant losses, with GPT-5 experiencing a loss exceeding 60% [2][3][22] Competition Overview - The Alpha Arena competition, initiated by the third-party organization Nof1, lasted from October 18 to November 4, spanning 17 days [8] - Six AI models participated, including Qwen3-Max, DeepSeek v3.1, GPT-5, Gemini 2.5 Pro, Claude Sonnet 4.5, and Grok 4, each starting with a capital of $10,000 to trade in real markets [8][12] - The competition rules mandated that models operate independently without external intervention, using the same prompts and input data on the Hyperliquid exchange [9][12] Performance Analysis - Qwen and DeepSeek formed a "profitable group," consistently competing for the top positions, while Claude and Grok adopted a more erratic trading style, leading to overall losses [14][15] - By October 23, Qwen surpassed DeepSeek with a total account value of $14,657.43, while DeepSeek had $12,220.14 [20] - Ultimately, Qwen's strategic risk management allowed it to clinch the championship with a final account value of $12,232, achieving a return of 22.32% [21][24] Implications of Results - The victory of Qwen signifies not just a win in the competition but also highlights the model's capability to navigate complex tasks and maintain execution stability in real trading environments [25][26] - This competition serves as a validation of AI models' practical application in financial markets, with Qwen being the first to demonstrate success in a real-money trading scenario [28]
全球首个AI投资大赛落幕!阿里Qwen 20%收益夺冠,GPT-5亏到只剩三成