Artificial-intelligence chatbots have seemingly taken the world by storm. But can they trade like human pros?
Not yet, judging by Alpha Arena, a new competition measuring the investing ability of six of the best-known large language models. Nof1, an AI start-up running the contest, gave each of the LLMs $10,000 each on Oct. 17, and told them to invest in cryptocurrency futures.
By this past Thursday, OpenAI’s ChatGPT held a position worth just $2,752, a 72% drawdown in six days, and Google’s Gemini tumbled 60%. Grok 4, run by Elon Musk’s xAI, and Anthropic’s Claude Sonnet also took losses. The only two LLMs in the green thus far are Alibaba’s Qwen3 Max and DeepSeek Chat, up 25% and 17%, respectively.
Of course, cryptos are volatile assets. But it shouldn’t be a surprise that chatbots aren’t yet good traders—they run on words, not numbers. “Large language models are trained on historical data, they’re backward-looking, and to win in markets you have to be adaptive,” Nof1 CEO Jay Azhang tells Barron’s. Training LLMs to trade is “a little like trying to get a goldfish to continue its train of thought.”
He hopes Nof1 can solve the issue. Nof1 is building a model trained on numbers, which it hopes will outperform the best investors. It set up Alpha Arena, which ends on Nov. 3, to create a benchmark for how existing chatbots trade. The Street’s top traders look safe for now.