Data from blockchain analysis platform CoinGlass shows that Chinese artificial intelligence models are outperforming US ones in crypto trading as competition intensifies among major generative AI chatbots.
AI chatbots DeepSeek and Qwen3 Max, both developed in China, led an ongoing cryptocurrency trading experiment on Wednesday, with the former becoming the only AI model to generate positive unrealized gains of 9.1%.
According to blockchain data platform Coinglass, the AI model “Qwen3” developed by Alibaba Cloud came in second place with an unrealized loss of 0.5%, followed by “Grok” with an unrealized loss of 1.24%.
OpenAI’s ChatGPT-5 fell to the bottom with a loss of over 66%, going from an initial account value of $10,000 to just $3,453 as of this writing.
The results surprised crypto traders, given that DeepSeek was developed at a fraction of the cost of its US rivals.
DeepSeek’s success came from betting on the rise of the cryptocurrency market. The model leveraged long positions across major cryptocurrencies such as Bitcoin (BTC), Ether (ETH), Solana (SOL), BNB (BNB), Dogecoin (DOGE), and XRP (XRP).
Related: Arthur Hayes asks for $1 million in Bitcoin as Japan’s new prime minister orders economic stimulus
DeepSeek outperforms all AI models with just $5.3 million in training funding
According to the model’s technical documentation, DeepSeek was developed at a total training cost of $5.3 million.
In comparison, OpenAI has reached a valuation of $500 billion, making it the world’s largest startup, Cointelegraph reported on October 2nd. According to corporate database platform Tracxn, the company has raised a total of $57 billion worth of capital across 11 funding rounds.
Although exact numbers for ChatGPT-5’s training budget have not been made public, Reuters reported in September that OpenAI spent $5.7 billion on research and development initiatives in the first half of 2025 alone.
According to a May 2024 X post by certified financial analyst Vladimir Kiselev, the total training cost for ChatGPT-5 is estimated to be between $1.7 billion and $2.5 billion.
Related: $19 billion market crash paves the way for Bitcoin’s rise to $200,000: Standard Chartered
Discrepancies in AI model cryptocurrency trading may be due to training data: Nansen analyst
Nikolaj Sondergaard, a research analyst at Nansen, a cryptocurrency intelligence platform, said differences in the cryptocurrency trading performance of AI models are likely due to the training data.
While ChatGPT is a good “general purpose” large-scale language model (LLM), another AI model, Claude, is primarily used for coding, an analyst told Cointelegraph, adding:
“If you look at historical income statements over the years, the models typically have very large price fluctuations, such as an increase in value of $3,000 to $4,000, but then a bad trade, or the LLM exits the trade because it gets caught in a big move.”
The performance of some of these AI models, particularly ChatGPT and Google’s Gemini, could be improved with the right prompts, said Kasper Vanderok, a strategic advisor and former quantitative trader.
“Maybe ChatGPT and Gemini could be better with different prompts. LLM is all about prompts, so it might probably perform worse by default,” Vandeloock told Cointelegraph.
While AI tools can help day traders identify changes in market trends through social media and technical signals, traders still cannot rely on AI tools for autonomous trading.
The competition began with a starting capital of $200 for each bot, which was later increased to $10,000 per model with trades executed on decentralized exchange Hyperliquid.
Magazine: Cryptocurrency traders are ‘fooling themselves’ with price predictions — Peter Brandt