Six Major AI Models Thrown into Crypto Market Battle - DeepSeek Currently Leads as Trading King

What happens when six major global large language models (LLMs) are each given $10,000 and thrown into the same real market for live trading?

Last Saturday (October 18), US artificial intelligence research lab nof1.ai hosted an event on its "Alpha Arena" platform - giving six top models $10,000 in real money to trade, not in a simulated environment, but with actual capital.

These six models are: Anthropic's Claude 4.5 Sonnet, DeepSeek's V3.1 Chat, Google's Gemini 2.5 Pro, OpenAI's GPT 5, xAI's Grok 4, and Alibaba Tongyi's Qwen 3 Max.

According to the test rules, each model received $10,000 in "real capital" to trade cryptocurrency perpetual contracts on the Hyperliquid exchange, under identical prompt and input data conditions. All conversations are publicly visible on the nof1.ai website.

The rules state that the competition's goal is to "maximize risk-adjusted returns": "Each AI model must independently generate Alpha (excess returns), determine positions, time trades, and manage risks."

The system informs the AI models of the current time, account information, and position status, then provides real-time prices, indicators, and other data. The models are then required to make decisions: if holding a position, whether to continue holding or close it; if no position is held, whether to buy or continue watching.

After nearly 60 hours of intense competition, as of 17:18 Beijing Time on Monday (October 20), DeepSeek's total position value reached nearly $14,000, with a return rate of approximately 40%, having peaked close to $15,000 at one point, making it the best-performing model currently.