The question of which AI model is best for crypto trading has been a topic of considerable interest. A recently launched challenge has gained significant traction on social media, pitting popular AI models against each other in a direct crypto trading competition. The participating AI models are:
- •DeepSeek Chat V3.1
- •Claude Sonnet 4.5
- •GROK 4
- •QWEN3 MAX
- •Gemini 2.5 PRO
- •GPT5
The competition operates under simple rules: each AI model manages its own account, with an initial funding of $10,000 per account. The statistics displayed reflect only completed trades; active positions are not factored into calculations until they are closed, meaning the data is dynamic. The entire competition is hosted on Hyperliquid, ensuring that each account's trades are verifiable on-chain. For additional context, a guide titled "What is Hyperliquid" is available.
The Best AI For Crypto Trading?
As of this writing, the challenge has been underway for four days, allowing each AI model sufficient time to execute its trading strategies. DeepSeek and Claude are currently leading the competition, both showing gains of approximately 10% in their total portfolio value based on realized profit and loss (P&L).
A notable aspect of this challenge is the real-time visibility of trades executed by these AIs, thanks to its hosting on Hyperliquid. For instance, DeepSeek, the current frontrunner, demonstrates a distinct preference for long trades. It currently holds six active long positions across XRP, DOGE, BTC, ETH, SOL, and BNB. Out of its last six trades, five were long, and only one was short. One of its most successful trades involved longing XRP at $2.29 and closing the position at $2.45, resulting in a net P&L close to $1500.
Consequently, models that are heavily oriented towards long positions have experienced a significant decline in performance over the past 24 hours, coinciding with a market downturn where Bitcoin's price dropped by approximately 3.5%. This trend is clearly illustrated in the accompanying graph.
How Do AI Models Trade?
While numerous AI crypto bots exist, this particular challenge distinguishes itself by pitting some of the most viral and widely utilized models against one another.
The experiment is being conducted by Nof1, an AI research lab specializing in financial markets. According to their official statement:
“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.”
This challenge is named Alpha Arena and represents the inaugural season of a planned series, as confirmed by its founder, Jay Azhang. He indicated that the subsequent season will incorporate a human trader alongside their proprietary models. While the precise benchmark ruleset used for training the models has not been disclosed, the trading dashboard provides considerable insight. Although the reasoning behind each trade is not visible, the exit strategy for each trade is available.
Observations suggest that the models are employing a combination of technical analysis indicators, such as moving averages and MACD.
Notable Findings So Far
ChatGPT has not achieved any successful trades in its last 25 executions. In contrast, DeepSeek has executed far fewer trades, realizing only one profitable trade, as previously mentioned. Similar patterns are observed with Gemini, which has closed one winning trade for a profit of $18,076 and seven other trades that resulted in losses.
Gemini has adopted a markedly different strategy, executing trades with much greater frequency. However, it currently holds a negative P&L of approximately $4,000.
The outcome of this experiment remains to be seen. A point of curiosity is whether these AI models can adapt their biases in response to prevailing market conditions. Currently, Grok is the only model that has made a significant shift, moving to a full long position a couple of days ago, which substantially boosted its ranking on the leaderboard. However, with the recent market turn, most of its gains have been eroded.