- DeepSeek was the only AI model to post a positive return in CoinGlass’ crypto trading experiment.
- Developed with just $5.3 million, DeepSeek outperformed OpenAI’s ChatGPT-5, which saw a 66% loss.
- Analysts say training data and prompt quality may explain the performance gap.
Chinese AI models are outperforming their U.S. counterparts in cryptocurrency trading, according to data from blockchain analytics platform CoinGlass. AI chatbots DeepSeek and Qwen3 Max, both developed in China, led the ongoing crypto trading experiment on Wednesday, with DeepSeek standing out as the only model to generate a positive unrealized return of 9.1%.
Qwen3 Max, developed by Alibaba Cloud, followed with a 0.5% unrealized loss, while Elon Musk’s Grok recorded a 1.24% loss. OpenAI’s ChatGPT-5 trailed with a steep 66% decline, bringing its initial $10,000 account value down to just $3,453.
DeepSeek achieves top results with $5.3 million in training costs
Despite being trained on one of the smallest budgets among its peers, DeepSeek achieved the best performance. Developed at a total cost of $5.3 million, it trailed far behind OpenAI’s multi-billion-dollar research expenditures.
Newsletter
Get weekly updates on the newest crypto stories, case studies and tips right in your mailbox.
DeepSeek’s trading strategy focused on leveraged long positions across major cryptocurrencies including Bitcoin, Ether, Solana, BNB, Dogecoin, and XRP — successfully capturing upside movements in the crypto market.
By comparison, OpenAI’s ChatGPT-5 was developed as part of a company valued at $500 billion, with estimated training costs between $1.7 billion and $2.5 billion.
Analysts point to data and prompts as key differentiators
According to Nansen research analyst Nicolai Sondergaard, the performance gap likely reflects differences in training data and model optimization. While ChatGPT remains a general-purpose language model, other AIs like DeepSeek and Claude may have been fine-tuned with specialized market data.
Former quantitative trader Kasper Vandeloock added that prompt design plays a major role in trading results: LLMs are all about the prompt maybe ChatGPT and Gemini could perform better with improved inputs.
Although AI tools are increasingly used for identifying short-term trends and sentiment shifts, experts caution that autonomous trading remains unreliable. CoinGlass’ experiment, which began with $200 and later scaled to $10,000 per bot, continues to highlight both the potential and the limitations of AI-driven trading.
