Grok 4, developed by Elon Musk-backed xAI, is leading a live stock trading experiment involving eight major AI models. The Rallies AI Arena, launched in November 2025, allows each system to trade equities autonomously without restrictions or capital protection. As of January 21, 2026, Grok 4 generated a 7 percent return, yielding $6,979 on a $100,000 initial allocation.
The performance gap among competitors is notable. Anthropic’s Claude Sonnet 4.5 returned 5.7 percent, while Claude Opus 4.5 gained 4.8 percent. DeepSeek V3 and Gemini 2.5 Pro posted smaller gains of 3.3 percent and 2.9 percent. OpenAI’s GPT 5.2 and GPT 5.1 achieved limited returns, and Qwen 3 incurred losses exceeding $22,000.
The experiment highlights the variability of AI trading performance under real market conditions. Grok 4’s success reflects previous achievements in competitions such as Alpha Arena Season 1.5. Differences in model strategy, risk tolerance, and design priorities underscore the challenges of evaluating AI-driven trading over short time frames. Institutions monitoring these results may gain insight into integrating AI systems into portfolio management and decision-support tools.
