According to data from the blockchain analysis platform CoinGlass, Chinese artificial intelligence models have outperformed their American counterparts in cryptocurrency trading, coinciding with intensified competition among leading generative AI chatbots.
The AI chatbots developed in China, DeepSeek and Qwen3 Max, led the cryptocurrency trading experiment conducted on Wednesday, with the former being the only AI model to generate a 9.1% positive unrealized return.
According to blockchain data platform CoinGlass, the AI model Qwen3 developed by Alibaba Cloud ranked second with a 0.5% unrealized loss, followed closely by Grok with a 1.24% unrealized loss.
OpenAI's ChatGPT-5 fell to last place, with losses exceeding 66%. At the time of writing, its initial account value had dropped from $10,000 to just $3,453.
Given that DeepSeek's development cost is only a fraction of its American competitors, this result has surprised cryptocurrency traders.
DeepSeek's success stems from betting on a rise in the cryptocurrency market. The model took leveraged long positions on major cryptocurrencies such as Bitcoin (BTC), Ethereum (ETH), Solana (SOL), BNB (BNB), Dogecoin (DOGE), and Ripple (XRP).
According to the model's technical paper, DeepSeek's total training cost was $5.3 million.
In contrast, according to a report by Cointelegraph on October 2, OpenAI has reached a valuation of $500 billion, making it the largest startup in the world. According to data from the company database platform Tracxn, the company has raised a total of $57 billion in 11 funding rounds.
While the exact figure for ChatGPT-5's training budget has not been disclosed, Reuters reported in September that OpenAI spent $5.7 billion on R&D programs in the first half of 2025 alone.
According to a post by chartered financial analyst Vladimir Kiselev on X in May 2024, the total training cost of ChatGPT-5 is estimated to be between $1.7 billion and $2.5 billion.
According to Nicolai Sondergaard, a research analyst at the cryptocurrency intelligence platform Nansen, the differences in performance of AI models in cryptocurrency trading may stem from their training data.
The analyst told Cointelegraph that while ChatGPT is an excellent "general" large language model (LLM), Claude—another AI model—is primarily used for programming. He added:
According to strategic consultant and former quantitative trader Kasper Vandeloock, the performance of some of these AI models can also be improved with the right prompts, especially ChatGPT and Google's Gemini.
"Perhaps ChatGPT and Gemini would perform better under different prompts; LLMs are entirely dependent on prompts, so maybe they perform poorly by default," Vandeloock told Cointelegraph.
While AI tools can help day traders identify market trend shifts through social media and technical signals, traders still cannot rely on them for autonomous trading.
The competition started with each robot having an initial capital of $200, which was later increased to $10,000 per model, with trading executed on the decentralized exchange Hyperliquid.
Related: Hong Kong first to approve the first spot Solana (SOL) ETF, leading the U.S.
Original article: “DeepSeek and Qwen outpace ChatGPT and Grok, leading in AI autonomous cryptocurrency trading experiment”
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
