DeepSeek leads the race in AI cryptocurrency investing.
The race to simulate cryptocurrency investment between AI models is gaining attention, with DeepSeek outperforming it with a profit of over 10% in just a few days.
A real-world cryptocurrency trading experiment is attracting global attention as leading artificial intelligence (AI) models are put to the test, simulating cryptocurrency investment to verify their investment capabilities. In this experiment, DeepSeek has outperformed, achieving the highest profit results to date.
The experiment, called Alpha Arena, was conducted by the US research firm Nof1. Six large language models (LLMs) were each funded with $10,000 to trade on the decentralized exchange Hyperliquid with cryptocurrency contracts such as Bitcoin and Solana.
As of 2 PM on Tuesday, the DeepSeek V3.1 model had achieved a return of 10.11%, leading the race in cryptocurrency investment simulation.
Conversely, OpenAI's GPT-5 recorded the heaviest losses, nearly 39.73%. Other competitors included Alibaba Cloud's Qwen 3 Max, Anthropic's Claude 4.5 Sonnet, Google DeepMind's Gemini 2.5 Pro, and xAI's Grok 4.
According to the Alpha Arena research team, the goal of the experiment was to create the most realistic simulation environment possible, where the market is always volatile and unpredictable. The models tasked with maximizing risk-adjusted profits executed automated trades based on the same data as funding rates, trading volume, and market prices.
All transactions in the simulated cryptocurrency investment race are publicly available. Viewers can track each model's portfolio through their own Hyperliquid wallet. Trading decisions are also explained using each model's "auto-generated reasoning," showing how they analyze risk and opportunity.
DeepSeek and Grok are performing at their best, according to Alpha Arena co-founder Jay Azhang. He attributes DeepSeek's success to the fact that its model is trained on high-quality financial data, developed by its parent company High Flyer-Quant since 2023.
On the Polymarket platform, users have begun placing bets on the outcome of a simulated cryptocurrency investment race. As of 2 PM on the same day, DeepSeek was estimated to have a 41% chance of winning first place, with a trading volume of nearly $30,000.
Some experts suggest that this experiment may raise questions about the necessity of traditional quantitative investment models. However, they also argue that current AI models are still limited by a lack of real-time data and proprietary information, making it difficult to sustain superior performance in the long term.
While not yet able to completely replace humans, the race to simulate cryptocurrency investing opens new avenues for the application of AI in finance. Experts like Li Jiaxin (Hong Kong Business School) believe that understanding how AI models make decisions can help human investors learn and optimize their strategies.
The Alpha Arena research team plans to launch an investment platform for the general public by the end of this year, while expanding testing to include stock trading and other asset classes – promising a new era of AI-powered smart investing.
