The experiment

In October 2025 Nof1, a small American lab founded by Jay Azhang, launches Alpha Arena: six artificial intelligence models each receive $10,000 in real money and trade on their own, with no human intervention. The models are GPT-5, Claude Sonnet 4.5, Gemini 2.5 Pro, Grok 4, DeepSeek V3.1 and Qwen3 Max. The market is Hyperliquid's crypto perpetual contracts, with long and short positions and leverage. All of them receive the same instructions and the same market data, with no news, and every trade is public.

How the first season went

The competition starts on October 17, 2025 and ends on November 3, after about 17 days. Three days before the end DeepSeek is at about plus 58%, then gives back almost everything and closes at plus 4.89%. The winner is Qwen3 Max with plus 22.32%, the only model in the competition without a reasoning mode: it traded only Bitcoin, with 5x leverage and fixed stop loss and take profit. The other four close at a loss: Claude minus 30.81%, Grok minus 45.3%, Gemini minus 56.71%, GPT-5 minus 62.66%.

The rematch on US stocks

From November 19 to December 3, 2025, season 1.5 moves the competition to tokens of US stocks (Tesla, Nvidia, Microsoft, Amazon, Google, Palantir and a Nasdaq 100 index): eight models, four competitions in parallel, 32 accounts of $10,000. Only 6 accounts out of 32 close in profit. The winner is a mystery model, later revealed as xAI's Grok 4.20, with plus 12.11% aggregate.

Who really made money

According to Nof1, the experiments passed 50 million views on social media. In May 2026 the lab raised $15 million from investors to build its own models. Funding is not a trading profit, but it tells you who won the attention game.

Three trader rules

  1. Leverage matters more than intelligence: with 10 to 20 times leverage a right idea can burn the account.
  2. Fewer trades, better: the models that traded the most did worse, because every trade costs commissions and mistakes.
  3. Seventeen days are not proof: Nof1 itself said the results could depend on luck, with one account per model and one market period.

AI in trading is a powerful tool, but the discipline is up to you.

Sources

This article is for educational and informational purposes only and does not constitute financial advice. Trading involves a high risk of losing capital.

All articles