AI Models Compete in Stock Trading: DeepSeek V3.1 Leads Ahead of GPT and Gemini
In the new season of the Alpha Arena benchmark, leading AI models trade real stocks with real money, starting with a $10,000 deposit. At the beginning of the competition, China’s DeepSeek V3.1 took the lead, surpassing its Western rivals — Gemini 3 Pro and GPT-5.1.
The goal of the tournament is to assess the ability of AI to independently make decisions on the stock market across four disciplines, including conservative trading (Monk Mode) and high-risk trading (Max Leverage).
According to the current leaderboard, DeepSeek V3.1 holds first place overall. The model uses the conservative Monk Mode, where it has already increased its capital to $15,000. This confirms an emerging trend: Chinese models are demonstrating higher profitability compared to GPT-5.1 and Gemini 3 Pro — a pattern also seen in previous crypto-market rounds.
DeepSeek and Kimi K2-Thinking show the strongest results in total returns and Sharpe ratio. Meanwhile, strategies using aggressive leverage are already producing losses.
The benchmark will continue for nearly two more weeks.
ORIENT
