Four of the most popular AI programs crashed and burned in a cryptocurrency investing competition, with most losing more than 50% of the money they were asked to maximize.
Only two Chinese programs managed to make a modest profit.
empiricism Alpha Arena Nof1’s competition awarded six AI models $10,000, matching input data, and prompted each of them to make as much money as possible while trading cryptocurrency stocks on the open market from October 17 to November 3.
Goal: Find the robot that can best make you rich.
Jay Azhang, 34, told The Post that the goal of his brainchild project — and future iterations of the contest — is to help the average person get the best tools for making money.
“Our goal is to help people trade better and give them access, hopefully, to the latest in technology when it comes to trading models,” Azhang said. “It is a unique field of research.
“Nobody really trains these models to look at all these numbers.”
Grok AI software from xAI; Anthropic Claude Sonnet; Google Gemini. OpenAI’s ChatGPT and Chinese-owned bots Deepseek and Qwen were given five-figure endowments and were instructed to make all other decisions — including whether to bet long or short their choice of cryptocurrencies.
The bots have been able to invest in cryptocurrencies such as Bitcoin,
Azhang said the cryptocurrency market, rather than trading stocks and shares, has proven to be better by competitive standards due to readily available blockchain data, 24-hour trading cycle, and lack of influence and leverage that retail traders compare to hedge funds in other financial markets.
“There is also more volatility,” Azhang said of the cryptocurrency market. “So a little more exciting.”
The results showed shocking inefficiency.
ChatGPT, the most popular bot according to StatCounterfinishing with just $3,794 – down 63%.
The worst of all was Gemini, which was down 56% with just $4,485 remaining, despite making the largest number of trades, 272.
Elon Musk’s Grok and Anthropic’s Claude Sonnet both achieved average results and were occasionally profitable during the 17-day trial. But Grok finished down 45% with $5,226, while Claude Sonnet took in $6,740, down 30%.
Chinese model Deepseek saw profits dip in the final days of trading, but ended higher – with $10,476 for a modest 4% return. The percentage reached approximately 100% on October 26.
Cowen, from Chinese company Alibaba, was the most volatile — operating at a loss on his initial investment for the first three days, then managing to dump all but $90 of his remaining balance in a long position in Bitcoin. She won the competition with 20% growth and $12,287.
From October 17 to November 3, Bitcoin fell 0.44%, Ethereum fell 11%, XRP rose 0.87%, BNB fell 8%, Doge fell 10%, and Solana rose just over 1%, according to data from Coinbase.
“It’s hard to say how much we can extract from this,” Azhang said. “The only thing we know is that there are patterns in the models and they are clearly biased and have preferences.
“For example, Claude is always buying and refuses to sell short. He is like an eternal optimist while Gemini is happy to short sell,” Azhang said. “They clearly have these inductive biases when it comes to trading.”
Alpha Arena is planning a next round that will add more AI models and also task the software with trading stocks alongside cryptocurrencies.
“We’re just getting started,” Azhang told The Post.
