Chinese artificial intelligence models are demonstrating superior performance compared to their United States counterparts in the realm of cryptocurrency trading, according to recent data analysis from the blockchain analytics platform CoinGlass. This development emerges as competition intensifies among leading generative AI chatbots.
On Wednesday, AI models DeepSeek and Qwen3 Max, both developed in China, led the ongoing crypto trading experiment. DeepSeek was the sole AI model to achieve a positive unrealized return, registering 9.1%. Qwen3, an AI model from Alibaba Cloud, secured the second position with an unrealized loss of 0.5%. Following closely, Grok recorded an unrealized loss of 1.24%, as indicated by data from CoinGlass.
In contrast, OpenAI's ChatGPT-5 experienced a significant downturn, slipping to the last place with an unrealized loss exceeding 66%. Its initial account value of $10,000 was reduced to just $3,453 at the time of reporting. These results have garnered surprise among crypto traders, particularly given that DeepSeek was developed at a considerably lower cost than its US-based competitors.
DeepSeek's successful trading strategy involved making bullish bets on the cryptocurrency market. The model strategically took leveraged long positions across major cryptocurrencies, including Bitcoin (BTC), Ether (ETH), Solana (SOL), BNB (BNB), Dogecoin (DOGE), and XRP.
DeepSeek's Cost-Effective Development Contrasts with US AI Giants
DeepSeek's development was completed at a total training cost of $5.3 million, according to the model's technical paper. This figure stands in stark contrast to the substantial investments made by US AI companies.
For perspective, OpenAI recently achieved a valuation of $500 billion, positioning it as the world's largest startup, as reported by Cointelegraph on October 2. The company has secured a cumulative $57 billion in capital through 11 funding rounds, according to data from the company database platform Tracxn.
While precise figures for ChatGPT-5's training budget are not publicly disclosed, OpenAI reportedly spent $5.7 billion on research and development initiatives in the first half of 2025 alone, according to a September report by Reuters. Estimates from May 2024, shared by chartered financial analyst Vladimir Kiselev on X, suggest that ChatGPT-5's total training cost could range between $1.7 billion and $2.5 billion.
Training Data May Explain Performance Discrepancies in AI Crypto Trading
The notable differences in crypto trading performance among AI models may be attributed to their respective training data, according to Nicolai Sondergaard, a research analyst at the crypto intelligence platform Nansen.
Sondergaard explained that while ChatGPT excels as a "general-purpose" large language model (LLM), other models like Claude are primarily utilized for coding. He noted, "Looking over the historical PNLs so far, models generally have very large price swings, like being up $3,000 - $4,000 but then making a bad trade or getting caught on big moves, causing the LLM to close the trade."
The performance of some AI models could potentially be enhanced with optimized prompting strategies, particularly for ChatGPT and Google's Gemini. Kasper Vandeloock, a strategic adviser and former quantitative trader, commented, "Maybe ChatGPT & Gemini could be better with a different prompt; LLMs are all about the prompt, so maybe by default they perform worse."
Although AI tools can assist day traders in identifying market trend shifts through social media and technical signals, they are not yet reliable for autonomous trading. The competition, which initially began with $200 in starting capital for each bot, was later increased to $10,000 per model. Trades were executed on the decentralized exchange Hyperliquid.