The newly released data is a bit eye-opening: OpenAI, Anthropic, Google, xAI, Alibaba—five of the strongest AI models, each trading with $10k in real cash.

32 results, only 6 profitable. Win rate 18.75%.
Every day someone claims "AI will replace traders," but let's look at this number first.
This experiment is called Alpha Arena, conducted by a lab called Nof1. The rules are simple: put the top AI models on a stage, each with $10k, let them trade cryptocurrencies autonomously, and see who can make money.
The result: most of the time losing money, frequent trading eating up fees, risk control a complete mess.
Those chatty AI models that respond fluently are no different from novice traders in the real money market.
Why are they so bad? The Nof1 team’s explanation hits the nail on the head:
Current large models are not trained for "allocating funds," but for "predicting the next word."
Let it write code, articles, answer questions—it's trained on trillions of tokens.
Let it decide "whether to buy this candlestick"—it hasn't been specifically trained for that.
Let ChatGPT trade crypto, it's like asking a linguistics PhD to play in the NBA—completely different skills.
But here’s the interesting part—
This week, sui:native Group (a NASDAQ-listed company) and Karatage jointly invested $15 million into Nof1.
The gamble is: general AI might not cut it, but AI trained specifically for the market could succeed.
Whoever builds the "AI Market Training Ground" infrastructure first will hold the key to the next generation of finance.
This is the smart money’s play—it's not about chasing AI narratives, but betting on which AI can actually make money in real-world applications.
What can ordinary people take away?
Short term: Don’t trust any "AI automatic trading" products.
If they were profitable, they would have made a fortune quietly long ago; most of the products pushing them are just collecting your IQ tax.
Long term: AI + finance is a real trend, but it will take several more years to mature.
In these years, those who use AI for research and data analysis will earn more than those relying on "AI helps me place orders."
Tools are for assistance; judgment still depends on yourself.