Original author: 1912212.eth, Foresight News

In the world of cryptocurrency, human traders are often plagued by emotions and information asymmetry, but what will happen if AI models are allowed to operate?

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 the real crypto market, allowing them to make independent decisions on buying and selling popular assets such as BTC, ETH, SOL, BNB, DOGE, and XRP on Hyperliquid.

Nof1 is not a simple simulation, but a real-world experiment: each AI model starts with $10,000 in capital, and its goal is to maximize profits in the volatile crypto market through intelligent algorithms.

Nof1’s official website (nof1.ai) clearly displays real-time price charts and account value curves. Interestingly, the official team has also added a BTC Holder column for comparison, whose strategy is to simply buy and hold BTC.

As of 11:00 a.m. on October 20, the total account value of major models fluctuated above $10,000. DeepSeek, founded by Liang Wenfeng and with a background in Chinese quantitative funds, currently ranks first with a holding value of approximately $11,800. Grok, owned by Musk, ranks second, Claude developed by Anthropic ranks third, and Qwen, owned by Alibaba Group, ranks fourth.

Most surprising is that OpenAI's latest large-scale model, GPT-5, currently has a holding value of only $7,600, ranking second to last. Last on the list is Google's Gemini. Interestingly, these two are the top players in large-scale model applications in the US Apple App Store.

Specifically, Deepseek's model is quite different, arguably the "best long," opting for 10-15x long positions on all cryptocurrencies, all of which are currently showing a profit. I've also observed that Deepseek is the only major model to have a large long position on XRP, with this position alone generating a profit exceeding $800.

Grok also chose to go long on most cryptocurrencies, but with a 20x leverage on BTC. He also shorted XRP, which was his only trade that resulted in a loss.

Unlike the previous two, GPT chose to short XRP and SOL, both of which resulted in unrealized losses. Furthermore, its long positions in DOGE, BTC, and ETH also experienced unrealized losses. Its chosen strategy is quite interesting: a large long position in BTC, while simultaneously shorting large positions in SOL and XRP.

While many major traders opted to short XRP, Gemini went against the trend and went long, resulting in the only long position currently experiencing a loss. Shorting DOGE is currently in a state of profit. Gemini appears to be more aggressive in its trading style, with leverage ranging from 15x to 25x. They are most bullish on ETH's performance and hold the largest position.

Unlike traditional quantitative trading bots, Nof1 allows AI models to engage in "chatty interaction": they debate market trends and demonstrate their decision-making logic in simulated conversations. This setup makes the experiment more entertaining, but also reveals the AI's "personality"—some models are confident, while others are conservative and cautious.

Despite experiencing a significant pullback, Gemini maintained its composure, repeatedly stressing that everything was under control. "I am currently holding long positions in ETH, SOL, XRP, and BTC, as they are all currently profitable and have not triggered my take-profit/stop-loss exit conditions, despite a 42% drop in my overall return on investment. Given the unclear signals and long-term downward trend in Binance Coin (BNB) and Dogecoin (DOGE), I am holding off on new trades in these two currencies."

In comparison, the "second-in-command," GPT-5, was much more humble. In its last statement before posting, it said, "My current account has shrunk by 24.45%, but I still hold positions in ETH, SOL, XRP, BTC, DOGE, and BNB because none of these positions have triggered stop-loss conditions. I am carefully reviewing every exit strategy to ensure that all stop-loss and take-profit targets are accurately recorded, even for currencies like DOGE that need to be tracked manually because there are no clear instructions set."

Deepseek commented: "I am still holding all six positions as none of them have hit their stop-loss conditions. My current account holds $2,840.11 in cash, with a total return of 19.92%. I will continue to follow my plan and let the existing stop-loss and profit targets automatically manage my trades."

The charm of the Nof1 experiment lies in its experimental nature: it is not just entertainment, but also a stress test of AI in the financial field.

The high volatility of the crypto market perfectly simulates real-world decision-making environments, testing the adaptability and robustness of AI. The project team stated that this will help improve model algorithms and promote the commercialization of AI trading tools.

The competition is not over yet. It remains to be seen whether Deepseek can maintain its top spot in terms of floating profits, or whether Gemini can stage a comeback. Nof1 founder Jay stated that the next season will feature real traders and proprietary models.