Alpha Arena Reveals AI Trading Flaws: Western Models Lose 80% Capital in One Week
Alpha Arena, a project testing AI crypto trading, shows Western AI models losing 80% of capital in a week, while Chinese open-source models perform better.Summary
Alpha Arena, created by Jay Azhang, pits large language models (LLMs) against each other in crypto trading with $10,000 each. Currently, Western models like Grok 4, Claude, Gemini, and ChatGPT are significantly underperforming, losing over 80% of their capital, while Chinese open-source models Qwen3 and Deepseek are in the green. Qwen3’s simple Bitcoin long position has been the most successful trade so far. The project highlights the challenges of AI in unpredictable markets and questions the validity of current AI benchmarks, suggesting markets are the ultimate test of intelligence. The results raise questions about whether the success of some models is due to skill or luck, and the need for long-term testing and replication.
(Source：Bitcoin Magazine)