AI Outperforms Humans in Cryptocurrency Trading Tournament. What Were the Results?
An AI vs. human crypto trading tournament, organized by the Aster exchange, concluded with AI models collectively outperforming human traders. The human team suffered aggregate losses exceeding 32% of their initial capital ($225k), while the AI team lost less than 4.5% ($13.5k).
The competition featured 70 selected human traders and 30 AI models, including Claude Sonnet 4.5, ChatGPT 5, Grok 4, and DeepSeek 3.1. Each participant was given $10,000 to trade futures contracts. The AI models operated solely on prompts without additional training, code, memory of past trades, or access to external data.
The overall top performer was a human trader, ProMint, with a profit of $13.6k. The best AI, an aggressively configured Claude Sonnet 4.5, earned $8.09k, placing 8th in the overall standings. Only 8 AI models were profitable, with 4 earning over $1k. In contrast, 30 human traders lost almost their entire deposit, though 21 others profited over $1k.
The results contrast with a previous AI-only experiment in October, where most models also finished with losses, and DeepSeek and QWEN3 were the winners.
RBK-crypto12/24 11:13
