Claude vs ChatGPT vs Grok in Trading: Who Made Money in Public Tests?
Admiral Markets
Picture: Admiral Markets
Public trading tests of large language models measured one thing. They showed how each model handled a small real-money account over a few weeks. Every model got the same prompt and the same data. They left out a great deal. Statistical significance was missing. Other markets and newer model versions were also outside the tests. The best-known test is Nof1's Alpha Arena. In Season 1 (crypto, October to November 2025), only two of six models made money, and both came from Chinese developers, according to Protos. In Season 1.5 (US equity exposure, November to December 2025), an unreleased xAI build was the only profitable model, according to the organiser's archived page. These are past results of short experiments. Past performance is not a reliable indicator of future results.
This report was first published by Admiral Markets. What you see here is a brief excerpt. The full text is theirs.
Continue reading at Admiral Markets
