Two Chinese artificial intelligence chatbots have demonstrated superior performance in an autonomous cryptocurrency trading competition, surpassing even advanced models like OpenAI’s ChatGPT. The competition concluded on Tuesday, revealing the unexpected success of budget AI models QWEN3 MAX and DeepSeek, which secured first and second place, respectively.
QWEN3 emerged as the sole AI chatbot to achieve positive returns, generating a total profit of $751 with a 7.5% return rate. In contrast, all other participating AI bots concluded the competition with losses, according to data from aggregator CoinGlass.
OpenAI’s ChatGPT finished at the bottom of the rankings, experiencing a significant 57% loss. Its initial investment of $10,000 was reduced to just $4,272 by the competition's end.
To achieve its winning position, QWEN3 employed a 20x leveraged long position on Bitcoin (BTC). The AI models were restricted to opening positions as of Wednesday.
According to CoinGlass data, QWEN3 initiated this leveraged bet when Bitcoin was trading at $104,556. The model faces liquidation if the price of BTC falls below $100,630.
Prior to the competition's conclusion, QWEN 3 predominantly held leveraged long positions on Bitcoin, Ether (ETH), and Dogecoin (DOGE).
ChatGPT's Underperformance in Crypto Trading
The results of this competition highlight that even the most heavily funded AI models still exhibit limitations in real-time crypto trading capabilities.
ChatGPT's last-place finish occurred despite OpenAI’s substantial investment in research and development. In the first half of 2025 alone, OpenAI reportedly spent $5.7 billion on these initiatives, according to Reuters.
While the exact budget for QWEN3 is not publicly disclosed, estimates from machine learning engineer Aakarshit Srivastava suggest that its training costs may have ranged between $10 million and $20 million.
DeepSeek secured second place, with its total training cost estimated at $5.3 million, as detailed in the model’s technical paper.
The Alpha Arena competition commenced with each bot starting with $200 in capital, which was later increased to $10,000 per model. All trades were executed on the decentralized exchange Hyperliquid.