AI Models Struggle in Cryptocurrency Investment Contest
Four popular AI programs floundered in a recent cryptocurrency investment competition, with most of them losing over 50% of the initial funds they were tasked to grow.
Only two Chinese models managed to achieve even a small profit.
In the Nof1 competition, dubbed alpha arena, six AI models were each given $10,000 and identical data inputs, aimed at maximizing returns while trading cryptocurrencies from October 17 to November 3.
The objective? Determine which AI could generate the highest returns.
Jay Azan, the contest’s creator, shared his vision of helping everyday individuals access superior trading tools.
“Our aim is to enhance trading capabilities and provide cutting-edge models for this unique research field,” Azan explained. “No one is really training these models to understand all these numbers.”
Models included Grok from xAI, Anthropic’s Claude Sonnet, Google Gemini, OpenAI’s ChatGPT, and Chinese-owned bots Deepseek and Qwen, all of which received five-figure budgets to make independent trading choices.
They could invest in major cryptocurrencies like Bitcoin, Ethereum, and others, with transactions meticulously recorded and made public, allowing real-time tracking of each model’s performance.
Azan noted that the cryptocurrency market offers a better competitive environment than stock trading, thanks to continuously accessible blockchain data and 24/7 trading hours, minimizing the retail traders’ advantages observed in other markets.
“There’s also more volatility,” he added. “It keeps things a bit more thrilling.”
The results, however, were quite underwhelming.
ChatGPT, the most recognized bot, ended with just $3,794—a staggering 63% loss.
Next was Gemini, which dropped 56% to leave it with only $4,485, despite executing the most trades—272 in total.
Grok and Claude Sonnet fared somewhat better, experiencing minor gains over the 17 days but still finishing down; Grok dropped 45% to $5,226 while Claude Sonnet was down 30% at $6,740.
Chinese model Deepseek, while losing some profits on the last trading day, still emerged with a modest gain of 4%, finishing at $10,476. It had been up 100% around October 26.
The Alibaba-affiliated Qwen exhibited the highest volatility. After an unsteady start with losses in its initial days, it eventually took a long position on Bitcoin, achieving a 20% increase and closing at $12,287.
From October 17 to November 3, Bitcoin dipped 0.44%, Ethereum dropped 11%, XRP increased by 0.87%, while BNB, Doge, and others saw varying losses, as per Coinbase data.
“It’s challenging to quantify what we’ve learned,” Azan remarked. “We do know that these models have distinct patterns and biases.”
“For instance, Claude stays almost exclusively long on positions, embodying an eternal optimist, while Gemini is more flexible, willing to go short,” he noted. “Clearly, they exhibit certain trading biases.”
The Alpha Arena team plans to expand in the next round, introducing additional AI models and extending their focus to stock trading alongside cryptocurrencies.
“We’re just getting started,” Azan concluded.
