Chinese artificial intelligence models are surpassing their U.S. peers in crypto trading, according to data from blockchain analytics platform CoinGlass, highlighting the growing competition among generative AI chatbots.

Performance Overview

In a recent crypto trading experiment:

• DeepSeek (China) was the only model to generate a positive unrealized return of 9.1%, taking leveraged long positions across Bitcoin (BTC), Ether (ETH), Solana (SOL), BNB, Dogecoin (DOGE), and XRP.
• Qwen3 Max, developed by Alibaba Cloud, posted a 0.5% unrealized loss.
• Grok followed with a 1.24% unrealized loss.
• ChatGPT-5 (OpenAI) recorded a 66% loss, reducing an initial $10,000 account to $3,453.

The results surprised crypto traders, given that DeepSeek was developed at just $5.3 million, a fraction of the cost of U.S. rivals.

Training Costs Comparison

• DeepSeek: $5.3 million in total training capital
• ChatGPT-5: Estimated total training cost of $1.7–$2.5 billion, with OpenAI spending $5.7 billion on R&D in H1 2025 alone
• OpenAI’s valuation: $500 billion with $57 billion raised across 11 funding rounds

Reasons Behind Performance Gap

Analysts attribute the discrepancy to differences in training data and strategies:

• Nicolai Sondergaard (Nansen) noted that ChatGPT’s general-purpose LLM design makes it less effective in highly volatile markets, often closing trades prematurely after large price swings.
• Kasper Vandeloock, former quantitative trader, suggested that prompt design could significantly improve ChatGPT and Google’s Gemini trading performance.

The competition initially started with $200 in starting capital per bot, later increased to $10,000 per model, with trades executed on the decentralized exchange Hyperliquid.

Key Takeaways

• Chinese AI models, DeepSeek and Qwen3 Max, outperform U.S. rivals in crypto trading
• DeepSeek’s leveraged long strategy across multiple major cryptocurrencies drove its 9.1% gain
• Cost efficiency: DeepSeek trained for $5.3M vs. ChatGPT-5’s estimated $1.7–$2.5B
• Performance gaps stem from training data and prompt optimization, not necessarily model capability

While AI models show promise for spotting market trends and aiding traders, experts caution that autonomous crypto trading remains unreliable, and human oversight is still critical.

This experiment underscores the growing capabilities of Chinese AI in financial applications and the importance of training efficiency and data quality in achieving superior results.