Here's something you don't see every day: an AI model that actually made money trading stocks instead of just burning through compute costs.

Elon Musk's xAI says its Grok 4.20 model topped the leaderboard in Alpha Arena Season 1.5, a live trading competition where AI models from different companies compete head-to-head using real market data and identical starting capital. Grok turned $10,000 into approximately $11,060, posting returns between 10% and 12%.

The Only Profitable Bot in the Room

What makes this particularly interesting isn't just that Grok won—it's that it was apparently the only model to finish in the black at all. Entries from OpenAI, Alphabet Inc.'s (GOOGL) Google (GOOG), and other major AI developers all ended up losing money.

According to a post on X by the account X Freeze, Grok variants occupied four of the top six positions on the leaderboard. The model was initially listed as a "mystery model" during the competition, which tested multiple configurations including Situational Awareness, New Baseline, Max Leverage, and Monk Mode. The fact that different setups all performed well suggests the edge wasn't just lucky timing or a single clever strategy.

Paying for Those GPUs

Musk responded to the news with characteristic humor: "Ok, I think I see a way to pay for all those GPUs."

The comment points to a real tension in the AI industry right now. Companies are pouring billions into computing infrastructure to train and run these models, and the question of how they actually make money from them remains somewhat open. If an AI can consistently generate trading profits, that's one possible answer—though turning a 10% return in a single competition into a reliable revenue stream is obviously a much bigger challenge.

As X Freeze put it: "Grok 4.20 isn't just doing well on benchmarks. It's making real money in live markets." Whether that translates into a sustainable business model for xAI remains to be seen, but it's certainly gotten people's attention.