In an autonomous cryptocurrency trading competition organized by Alpha Arena, Chinese artificial intelligence models QWEN3 MAX and DeepSeek dominated the field, while OpenAI's ChatGPT finished last with significant losses. The competition, which began on October 18, initially provided each bot with $200 in capital, later increased to $10,000 per model, and all trades were executed on the decentralized exchange Hyperliquid.
QWEN3 MAX emerged as the only profitable participant, generating a 7.5% return rate with $751 in profit. It maintained leveraged long positions throughout, primarily focusing on Bitcoin (BTC), Ether (ETH), and Dogecoin (DOGE). By the competition's end on Tuesday, QWEN3 MAX held a 20x leveraged long position on Bitcoin, entered when BTC traded at $104,556, facing liquidation if the price drops below $100,630.
In contrast, ChatGPT recorded a 57% loss, reducing its $10,000 investment to just $4,272. This poor performance occurred despite OpenAI spending $5.7 billion on research and development in early 2025, with ChatGPT's training costs estimated between $10 million and $20 million. DeepSeek secured second place with a total training cost of only $5.3 million, highlighting the cost-efficiency of budget models. CoinGlass tracked all trading data, confirming these metrics and the surprising outcome that even well-funded AI models struggle with real-time crypto trading decisions.
