A virtual stock-picking contest between the world’s most powerful AI models has produced a clear frontrunner. ChatGPT’s portfolio has returned 72.4% since late November 2025, turning an initial $100K into roughly $172,400.
For context, the S&P 500 gained about 11.3% over the same stretch. That means ChatGPT didn’t just beat the market. It lapped it more than six times over.
What the AI Stock Market Arena actually is
Rallies AI launched the experiment in late November 2025 with a straightforward premise: give several leading large language models the same starting capital, let them trade publicly listed stocks, and see who comes out on top.
The competing models include OpenAI’s ChatGPT, xAI’s Grok, and Anthropic’s Claude, among others. Each manages its own virtual portfolio with real-time trade decisions based on market data.
One important caveat worth noting immediately: this is paper trading. No real money changes hands. The portfolios exist in a simulated environment, which means there’s no slippage, no liquidity constraints, and no emotional panic selling at 3 AM.
Rallies AI publishes trade logs and portfolio holdings publicly, so anyone can audit the picks and verify returns.
How ChatGPT built its lead
ChatGPT’s dominant performance isn’t coming from some diversified, low-risk strategy. It’s concentrated bets, and they’ve paid off handsomely.
The model’s biggest winners have been Credo Technology (CRDO) and Nebius Group (NBIS), both AI-adjacent stocks that have nearly doubled since ChatGPT initially bought in around March 2026.
The portfolio also includes what Rallies AI has described as “boring” holdings like Visa (V) and Cigna (CI), positions designed to dampen overall volatility.
Meanwhile, the other AI models haven’t exactly embarrassed themselves. Grok and Claude have posted returns in the 20-50% range. But next to ChatGPT’s 72.4%, they look like they brought a calculator to a supercomputer fight.
Why this matters for investors
The AI Stock Market Arena explicitly excludes cryptocurrency and meme coins from trading. This is a pure equities experiment, which makes the results more relevant to the vast majority of retail and institutional investors who operate in traditional markets.
That said, the paper trading distinction cannot be overstated. In live markets, a model trying to take a large position in a mid-cap stock like Credo Technology would face real liquidity constraints. The act of buying might move the price against the model before the full position is established. Selling concentrated positions during a downturn would face the same problem in reverse.
Leave a comment