In the Alpha Arena crypto trading competition, four out of six large language models (LLMs) ended up losing money, with ChatGPT suffering a 63% loss. The two-week contest saw only DeepSeek and QWEN3 MAX models achieving profitability. According to the organizer, Nof1, excessive trading costs significantly affected early outcomes, as models struggled with parsing market data and decision-making within limited context windows. A subsequent competition with enhanced prompts is scheduled to address these challenges.