Which AI model is best for crypto trading? Surely this is something on everyone’s mind for a while now. There have been plenty of people experimenting, but a recently-launched challenge has gone viral on social media, pinning the most popular models against each other in a straight-off crypto trading competition. The challengers are:
- DeepSeek Chat V3.1
- Claude Sonnet 4.5
- GROK 4
- QWEN3 MAX
- Gemini 2.5 PRO
- GPT5
The rules are simple: each AI model operates its account, each of which is funded with $10,000. All statistics we can see reflect completed trades only; active positions are not included in the calculations until they are closed, which means the data varies. The entire thing is hosted on Hyperliquid, meaning that each account’s trades are verifiable on-chain. Check out our guide what is Hyperliquid for more context.
The Best AI For Crypto Trading?
At the time of this writing, the challenge is four days old, meaning that all AI models have had this time to execute their strategies. Clearly in the lead are DeepSeek and Claude, both of which are up by around 10% on their total portfolio value in realized P&L.
The interesting bit about this is that since everything is hosted on Hyperliquid, we can see the trades these AIs are taking in real time. For example, at the time of this writing, DeepSeek, the current leader, has a clear bias towards long trades. Currently, it has six active long positions on XRP, DOGE, BTC, ETH, SOL, and BNB. In its last six trades, five were long, and only one was short. Its biggest winner was longing XRP at $2.29 and closing it at $2.45 for a net P&L of close to $1500.
Naturally, this also means that long-oriented models saw their performance decline considerably in the past 24 hours, as the market tumbled and BTC’s price dropped by some 3.5%. This is clearly visible in the graph that follows:
How do AI Models Trade?
While there are plenty of different AI crypto bots, this challenge is different in that it pits some of the most viral and widely used models against each other.
The team behind the experiment is Nof1 – an AI research lab focused on financial markets. According to their official page:
“At Nof1, we believe financial markets are the best training environment for the next era of AI. They are the ultimate world-modeling engine and the only benchmark that gets harder as AI gets smarter.”
The above challenge is called Alpha Arena and is the first season of what the team plans to do, as confirmed by their founder Jay Azhang, who recently said that the next season will include a human trader, as well as their “homegrown” models. So far, we haven’t found the exact benchmark ruleset that was used to train the models, but there is quite a bit of insight in the trading dashboard itself. For example, we can’t see the reasoning of each trade, but we can see the exit plan.
At a glance, it appears that the models are using a combination of various indicators, commonly used in technical analysis, such as moving averages and MACD.
Notable Findings so Far
ChatGPT hasn’t made one successful trade out of its last 25 executions. DeepSeek, on the other hand, has taken far fewer trades and has realized only one winning idea (as mentioned above). Similar observations can be seen with Gemini, which has closed one winning trade for a profit of $18076 and seven others for a loss.
Gemini has taken a completely different approach and executes trades a lot more frequently. However, it currently sits on a negative P&L of about $4000.
It’s interesting to see how this experiment will turn out. What’s curious is whether those AI models will be able to switch biases based on current market events. At the moment, the only one that did a major turnover was Grok, which went full long a couple of days ago and skyrocketed its position in the leaderboard. Since the market took a turn, though, most of its profits were wiped out.
The post Which AI is Best for Crypto Trading? Viral Challenge Puts ChatGPT, Grok, Claude, and More to the Test appeared first on CryptoPotato.
Source link
