In a recent cryptocurrency trading competition, AI models DeepSeek, Elon Musk’s Grok, and Claude Sonnet 4.5 demonstrated outstanding performance, gaining over 25% in profit. Meanwhile, flagship models from leading developers — GPT-5 (OpenAI) and Gemini 2.5 Pro (Google) — suffered significant losses exceeding 28%.
Each AI model received an initial capital of $10,000 to conduct independent futures trading. The models were tasked with generating trading ideas, determining trade volumes, and managing risk autonomously.
DeepSeek’s unexpected success suggests the advantage of specialized fine-tuning on financial datasets. At the same time, Grok’s strong results indicate that powerful general-purpose AI systems can also perform effectively in financial markets.
GPT-5, on the other hand, adopted a highly conservative strategy, making only a few small trades. This approach protected it from major drawdowns but prevented it from achieving notable profits.
Despite the evident potential of AI in analyzing large volumes of data, the severe losses of some models highlight the serious risks involved in real-world financial operations.
Key concerns for traders using AI include:
• Decision opacity (“Black box”): It is often impossible to understand how AI reaches critical financial conclusions.
• Tendency to “hallucinate”: The risk that a model may produce confident yet incorrect outputs, which can be disastrous in real trading conditions.
DeepSeek and Grok frequently rank among the leaders in such tests, proving that in the world of algorithmic trading, success does not always depend on the most famous developer’s name.
