AI companies’ cryptocurrency real-world investment battle\nDeepSeek’s highest return, 500% jackpot\nchatGPT, loss of thousands of 弗 in Gemini, ‘humiliating’\nDifference in ability to capture low points
![A phenomenal return on investment for 'Grook 4'. The Groc 4 (black line) had an overwhelming return of nearly 500%, with $200 in capital at the start of the tournament called more than $1100. This is the result of accurately capturing the market low and switching to a long position. [Source = Alpha Arena]](https://wimg.mk.co.kr/news/cms/202510/21/news-p.v1.20251021.5417aa38e1e14773bb0f94bd1afdb33c_P1.jpg)
It is a hot topic as Grok and DeepSeek overwhelmed big tech AI models such as ChatGPT and Gemini in the cryptocurrency real-world investment competition between generative artificial intelligence (AI).
In particular, xAI’s latest AI model “Grook 4” accurately captured the market’s short-term low and switched to a leveraged long position (buy), resulting in a phenomenal return of 500% on the first day of the competition alone.
According to blockchain data platform CoinGlass on the 21st, China’s DeepSeek was ranked as the most profitable chatbot with a total unrealized profit (PNL) of $3,650 (about 4.7 million won). Groch followed with an unrealized profit of about $3,000. Antropic’s “Claude Sonet 4.5” ($2340) and Alibaba’s “Qwen 3 Max” ($784) also performed well, recording profits.
On the other hand, the performance of OpenAI’s ChatGPT and Google’s Gemini, which ranked high in AI model performance measurements in other fields, was disastrous. OpenAI’s “Chat GPT 5” lost about $2,800 in unrealized losses, while Google’s “Gemini 2.5 Pro” lost face with a loss of as much as $3270. It is analyzed that these models maintained their initial short positions (sell) even at the time of market rebound, increasing their losses.
The competition, hosted by the developer community, drew attention as a test to gauge the actual investment capabilities of AI models. Each AI model started with an initial capital of $200 (then increased to $10,000) and proceeded by conducting actual transactions on the DEX Hyper Liquid.
![Thousands of dollars worth of unrealized gains and losses (PNL) trend recorded by top AI models at the Artificial Intelligence (AI) virtual asset real-world investment competition. [Source = Coin Glass]](https://wimg.mk.co.kr/news/cms/202510/21/news-p.v1.20251021.c0ec41250d674506bd18e08967688975_P1.png)
The highlight of this competition was the dramatic shift in position of ‘Grook 4’. “Grook accurately captured the local bottom of the market just before the recovery rally,” AI company founder Jaz Ajang Stealth said in his X. “We converted the existing short position to a leveraged long position and achieved a 500% portfolio return on the first day of the competition alone.”
Grok made aggressive long bets, using 20x leverage on Ripple (XRP), 15x on Solana (SOL) and 10x leverage on Dogecoin (DOGE). “The market rewards sharp timing and certainty,” Groch said. He also self-evaluated that he turned risk into profit.
Recently, a growing number of virtual asset traders are using AI chatbots to predict potential upside and downside markets or to understand when to buy or sell. This is because AI has strength in capturing changes in real-time investment sentiment such as social media public opinion.
However, an official from the virtual asset industry pointed out, “As seen in this competition, there was a big difference in market analysis capabilities for each AI model,” adding, “We can refer to AI’s insights, but it is too early to rely entirely on autonomous transactions.”