Claude Fabel 5 Agentic AI Trading: First Tests Looks VERY Strong
Why It Matters
This demonstrates how next-generation agentic LLMs can autonomously devise, execute and iteratively refine short-term trading strategies, suggesting faster strategy discovery and operational automation for quant trading—but also raising questions around risk controls, market impact and oversight.
Summary
Tester used Anthropic’s new Claude Fable 5 to run an agentic AI trading bot on five-minute Poly Markets data and reported a roughly $41 gain over ~10 hours with a 71% win rate (peak $82 in a day). Fable 5 autonomously analyzed historical snapshots, built a fee-aware fair-value signal and execution formula, and recommended rules including position sizing, hold resolution, and a 10% daily loss halt. The model also developed a mixed tactic that pairs conservative signals with occasional “deep long-shot fading” trades and the tester scheduled two-hour automated health checks to let the agent adjust strategy mid-run. Early live results impressed the tester and showed the model adapting its strategy based on observed trades.
Comments
Want to join the conversation?
Loading comments...