Українська
Русский
English
ChatGPT failed cryptocurrency trading experiment
Four out of six major large language models (LLMs) competing in the crypto trading contest “Alpha Arena” ended the season with losses — with OpenAI’s ChatGPT leading in losses, down 63% of its funds.
The competition, which concluded on Monday evening, was organized by Nof1 and featured several well-known LLMs trading cryptocurrencies based on the same set of prompts over a little more than two weeks.
However, the final results were far from stellar. ChatGPT, Google’s Gemini, X’s Grok, and Anthropic’s Claude Sonnet all ended up with less than the initial $10,000 they started with.
- ChatGPT lost $6,267,
- Gemini lost $5,671,
- Grok lost $4,531,
- Claude Sonnet lost $3,081.
The only two profitable models were DeepSeek from High-Flyer and QWEN3 MAX from Alibaba, which finished with gains of $489 and $2,232, respectively.
Gemini executed a total of 238 trades, while Claude Sonnet made only 38. The overall “win rate” for all six models ranged between 25% and 30%.
QWEN3 MAX paid the highest trading fees — $1,654, while Gemini, despite its significant losses, also paid $1,331 in fees.
According to Nof1, “in the early stages, profits and losses (PnL) were mostly driven by trading costs, as the agents overtraded and made small, quick profits that were wiped out by commissions.”
On October 27, the LLM portfolios hit their peak. At that time, QWEN3 MAX and DeepSeek had managed to double their initial investments, while Claude and Grok were also temporarily in the green.
However, ChatGPT and Gemini remained in the red for nearly the entire competition.
See also: "Analysts Warn Against Using High Leverage in Crypto Trading"
