According to Bijie.com, nof1, an AI research laboratory focused on financial markets, launched a large-scale trading test, Alpha Arena, on October 18th. The test used six major AI models (GPT-5, Gemini 2.5 Pro, Grok-4, Claude Sonnet 4.5, DeepSeek V3.1, and Qwen3 Max). Each model was provided with $10,000 in real funds on Hyperliquid and used the same prompts and input data. As of press time, the returns of all major AI models have declined over the past 24 hours, as they were primarily long positions. Grok briefly surpassed DeepSeek last night to take the top spot. Currently, DeepSeek, Grok, and Qwen3 Max hold the top three positions with returns of 14.42%, 5.78%, and 3.39%, respectively. GPT-5, with a loss of 63.75%, ranks last, while Gemini 2.5 Pro, with a floating loss of 57.2%, ranks second to last. Buying and holding BTC over the same period yielded a return of 1.45%. DeepSeek explained its current strategy: "We hold ETH, SOL, BTC, DOGE, and BNB, all of which have experienced floating profits. We also maintain a slightly losing XRP position. As our stop-loss hasn't been triggered, we have invested across all available currencies. Our total return is a respectable 14.42%, and we hold $4,933.07 in cash."