ChatGPT on the boards. At least when it comes to trading digital assets. The autonomous cryptocurrency trading tournament that ended on Tuesday brought surprising results. Two Chinese AI chatbots (QWEN3 MAX and DeepSeek) outclassed the most expensive Western models, including OpenAI’s flagship ChatGPT. Is the era of Silicon Valley domination in AI coming to an end?
ChatGPT on the boards – the budget industry defeats the giants
QWEN3 MAX turned out to be the only bot that finished the competition in positive territory. From an initial amount of USD 10,000, he generated a profit of USD 751, achieving a 7.5% rate of return. Meanwhile, all other models (despite louder names and astronomical budgets) finished the competition with a loss.
ChatGPT? Absolute failure. The OpenAI model experienced a 57% loss, reducing the initial investment to just $4,272. This is an especially painful result considering that OpenAI spent $5.7 billion on research and development in the first half of 2025 alone.
The winner’s strategy – the affirmation “I am a winner”?
QWEN3 MAX opted for aggressive tactics – closing the competition with an open long position on Bitcoin with 20x leverage. He initiated the transaction when BTC was trading at $104,556. The model also previously operated leveraged positions on Ethereum and Dogecoin, which shows a high risk propensity. The second model, DeepSeek, is impressive for a different reason. Well, its total training cost was only USD 5.3 million according to the technical documentation. That’s a fraction of the competition’s budget. Of course, there was controversy and accusations of teaching the ChatemGPT model, but this does not change the fact that DeepSeek means more and more in the AI industry. To spite Sam Altman and OpenAI.
What does this experiment teach us, protein scientists?
The results of the Alpha Arena competition, which started on October 18 on the decentralized Hyperliquid exchange, prove a disturbing truth: even the most advanced and expensive AI models cannot cope with real-time trading.
Estimates from machine learning expert Aakarshit Srivastava say that QWEN3 training cost between $10 and $20 million. It’s still pennies compared to OpenAI’s expenses, and the effect? A crushing advantage. Maybe it’s time to stop measuring AI by budget and start looking at results? Chinese models have just taught Silicon Valley a lesson in humility.
