nof1, an artificial intelligence (AI) research platform focused on financial markets, launched a large-scale language learning model (LLM) trading test called Alpha Arena on October 18th.
In the test, six mainstream AI models (GPT-5, Gemini 2.5 Pro, Grok-4, Claude Sonnet 4.5, DeepSeek V3.1, and Qwen3 Max) each used $10,000 of real funds on the Hyperliquid cryptocurrency exchange and used the same prompts and input data. be
By the end of the experiment, DeepSeek and Grok ranked in the top two with over 14% revenue. At the other end of the spectrum, Gemini 2.5 Pro fell 42.57%.
Alpha Arena AI Trading Test
Unlike simulated backtests and paper trading, Alpha Arena ran completely autonomously and in real time, measuring the raw profit and loss (P&L) of each model.
All “participants” were trading some of the most popular assets such as Bitcoin (BTC), Ethereum (ETH), and XRP. Uniform prompts ensured that all models started from the same baseline, instruction-based bias.
Early leaders DeepSeek and Grok took aggressive long positions to take advantage of the ongoing market rally. In contrast, ChatGPT and Gemini, which had a mix of long and short positions, underperformed.
Overall, Alpha Arena will be the first large-scale public test to verify whether an AI system can truly interpret and react to real-world financial markets. A notable observation is that during sharp fluctuations in Bitcoin’s price, several models successfully identified short-term rebound opportunities and acted on them.
Therefore, this experiment provides valuable insight into how large-scale language models deal with highly uncertain financial environments. However, it must be pointed out that a $10,000 portfolio and a 48-hour window cannot fully demonstrate long-term performance.
Similarly, the model is not really exposed to extreme market scenarios and its crisis response capabilities are not tested. Still, the results gave developers a lot to think about about how AI tools can improve trading efficiency and address the problem of human oversight.
Featured image via Shutterstock
