Original title: “Who is the best at AI cryptocurrency trading? Surprisingly, home-grown AI is taking the lead?”
Original author: 1912212.eth, Foresight News
In the world of cryptocurrencies, human traders often struggle with emotions and information asymmetries. But what happens if AI models take over the trading task?
On October 18, a project called Nof1 placed multiple AI models — GPT-5, Claude Sonnet 4.5, Gemini 2.5 Pro, Deepseek V3.1, and Qwen3 Max — in real-world cryptocurrency markets, allowing them to autonomously make long or short trades on popular assets like BTC, ETH, SOL, BNB, DOGE, and XRP on Hyperliquid.
Nof1 is not just a simulation but a live test with real money. Each AI model starts with $10,000, with the goal of maximizing profits through intelligent algorithms amid the highly volatile cryptocurrency market.
Nof1’s official website (nof1.ai) clearly displays real-time price charts and account value curves. Interestingly, the official team has added a section for Bitcoin holders to compare, using a buy and hold Bitcoin strategy.
As of 11:00 AM on October 20, the total account values of the various models have fluctuated above $10,000. DeepSeek, founded by Wenfeng Liang and backed by Chinese quant funds, currently ranks first with a portfolio value of around US$11,800. Grok, under the leadership of Elon Musk, is in second place, followed by Cloud, developed by Anthropic, in third place, and Coin, from Alibaba Group, in fourth place.
What’s even more surprising is that GPT-5, OpenAI’s latest big model, is currently worth just $7,600, ranking second to last, while Google’s Gemini is last. Interestingly, these two templates are among the top players in large template apps on the US Apple App Store.
Specifically, Deepseek’s approach is very distinctive and can be considered an “optimal long position” strategy. The model chose to go long on all cryptocurrencies with a leverage of 10-15 times, and at the moment, all positions are showing unrealized gains. It is worth noting that Deepseek is also the only large-scale model to have taken a significant long position in XRP, generating unrealized profits of over $800 on this position alone.
Likewise, Grok chooses to go long on most cryptocurrencies, but with a key difference: he uses 20x leverage on Bitcoin (BTC). Additionally, it took a short position on XRP. This short position on XRP is the only trade currently showing unrealized losses.
In contrast to the first two models, GPT chose to short XRP and SOL, both of which are currently showing unrealized losses. Moreover, its long positions in DOGE, BTC, and ETH also incurred unrealized losses. Its strategy is particularly interesting, as it heavily loaded Bitcoin while simultaneously taking large short positions on SOL and XRP.
While many other major models chose to sell XRP, Gemini took the opposite approach by continuing to buy XRP, resulting in its only unrealized loss on a long position. On the other hand, her short position on DOGE is currently profitable. Geminis seem to adopt a more aggressive trading style, with leverage ranging between 15 and 25 times. It is the most optimistic about ETH’s performance, as it has been assigned the heaviest position.
Unlike traditional quantitative trading bots, Nof1 allows AI models to engage in “interactive conversations”: they discuss market trends through simulated dialogues, and display their decision-making logic. This setup not only makes the experience more enjoyable, but also reveals the “personalities” of the AI models – some exude confidence, while others remain cautious and conservative.
Although Gemini witnessed a major decline, he excelled in his steadfastness, repeatedly confirming that everything is under control. “I currently hold long positions in ETH, SOL, The two digital currencies.
In contrast, the GPT-5 “Vice Captain” seemed more modest, stating in his last post before publication: “My current account has shrunk by 24.45%, but I continue to hold positions in ETH, SOL, Achieve all stop loss and take profit targets. “They are recorded accurately, even for DOGE, which requires manual tracking due to the lack of clear instructions.”
Deepseek commented: “I am still holding all 6 positions as none of them have triggered a stop loss condition. My account currently holds $2,840.11 in cash, and the total return rate is 19.92%. I will continue to proceed as planned, allowing my current stop loss levels and profit targets to manage the trades.” automatically.”
The magic of the N-of-1 experience lies in its experimental nature: it is not just entertainment but also a stress test for artificial intelligence in the financial field.
The reason is that the high volatility of the cryptocurrency market perfectly mimics real-world decision-making environments, testing the adaptability and power of artificial intelligence. The project team stated that this helps improve model algorithms and enhances the commercialization of AI trading tools.
Currently, the competition is not over yet. It is still uncertain whether Deepseek can maintain its top position in unrealized gains or whether Gemini can make a comeback. Jay, founder of N-of-1, stated that the next season will introduce human traders as well as self-developed models.
