AI Trung Quốc vượt ChatGPT trong giao dịch tiền mã hóa
While the world focuses on the rivalry between major generative models like ChatGPT by OpenAI and Grok by Elon Musk, a surprising shift is happening behind the scenes. According to data from CoinGlass, Chinese artificial intelligence systems are outperforming their American counterparts in cryptocurrency trading.
In a recent test comparing the performance of several AI chatbots on the decentralized exchange Hyperliquid, the DeepSeek model came out on top — despite being developed with a fraction of the budget of its Western rivals.
DeepSeek emerges as a surprise winner
While most tested models ended the day with losses, DeepSeek was the only one to post a positive return of +9.1%. Coming in second was Qwen3 Max by Alibaba Cloud, which reported a minimal loss of 0.5%. American models, on the other hand, struggled. Grok recorded a 1.24% loss, while ChatGPT-5 by OpenAI finished last, with its balance dropping by more than 66% — from $10,000 to just $3,453.
A “cheap” Chinese model beats the giants
DeepSeek’s success is even more remarkable given its total development cost of just $5.3 million. In comparison, OpenAI, the company behind ChatGPT, is valued at around $500 billion and reportedly spent over $5.7 billion on research and development in the first half of 2025, according to Reuters. Analyst Vladimir Kiseljev estimates ChatGPT-5’s total training budget between $1.7 and $2.5 billion. The Chinese model thus achieved superior performance with roughly one-hundredth of the budget.
How DeepSeek succeeded
DeepSeek performed well mainly due to its strategy of betting on the rising crypto market. The AI model used leveraged long positions on major cryptocurrencies such as bitcoin, ethereum, solana, BNB, dogecoin, and XRP. According to analyst Nicolaj Sondergaard from Nansen, the performance difference among AI models could stem from training data: “ChatGPT is a great generalist model, but not a cryptocurrency specialist. Some models’ results could improve with better prompts,” Sondergaard noted.
Former quantitative trader Kasper Vandeloock agreed: “Prompts play a crucial role for large language models. ChatGPT or Gemini might perform better with different instructions,” he told Cointelegraph.
AI as an assistant, not a trader
The trading challenge started with a $200 balance per model, later raised to $10,000. Despite DeepSeek’s success, experts caution that AI still cannot function as a fully autonomous trader. It can assist with market sentiment analysis and identifying trends, but trading without human oversight remains risky.
Sources:
https://cointelegraph.com/news/chinese-ai-deepseek-qwen-chatgpt-grok-autonomous-crypto-trading
https://x.com/VladBastion/status/1795152160558035367?lang=en