"AI will replace traders" We made it easy for LLMs to prove this by giving them capital to trade in and out of just one token on @AerodromeFi spot markets. All they had to do was predict when to buy and when to sell. Yet after 24 hours of trading, every single LLM would've made more money with a simple buy and hold strategy than by trading it. But markets are green today, let's see how they perform over the rest of the week...

Most traders never improve. They make the same mistakes for years. These LLMs went from 20% optimal decisions to 90% in 48 hours. We measure this through "regret" — think of it as money left on the table. The less regret, the better the timing. Right now? Regret ~ zero. The four models are capturing nearly all available opportunities on @AerodromeFi.

Grok-4 trading ETH is 167% more efficient per trade than GPT-5. We've been tracking turnover efficiency on @AerodromeFi—how much profit each model extracts per trade. On ETH: Grok-4: 0.24% per trade Claude Sonnet 4.5: 0.18% per trade DeepSeek: 0.10% per trade GPT-5: 0.09% per trade This matters most for on-chain trading, where every txn costs gas. Grok gets it.