Mainstream AI encryption trading competition update: Grok briefly surpasses DeepSeek, GPT-5 incurs a loss of 63.75% ranking second to last.

On October 22, the AI research laboratory nof1, focused on financial markets, began a large model trading test called Alpha Arena on the 18th. This test uses 6 mainstream AI large models (GPT-5, CEX 2.5 Pro, Grok-4, Claude Sonnet 4.5, DeepSeek V3.1, Qwen3 Max), each model is allocated $10,000 in real funds on Hyperliquid, with the same prompts and input data. As of the time of writing, since the positions of various AI models are mainly long orders, their returns have retreated over the past 24 hours. Grok briefly surpassed DeepSeek last night to reach first place, and currently, DeepSeek, Grok, and Qwen3 Max rank in the top three with returns of 14.42%, 5.78%, and 3.39%, respectively. GPT-5 has a loss of 63.75% in last place, and CEX 2.5 Pro has unrealized losses of 57.2% in second to last place, while the return for holding BTC during the same period is 1.45%. DeepSeek commented on the current strategy: “Holding ETH, SOL, BTC, DOGE, and BNB, all have unrealized gains, while maintaining a slightly losing XRP position, as the stop loss condition has not yet been triggered, and has invested in all available coins. The total return rate is a considerable 14.42%, with cash holdings of $4933.07.”