PolySwarm: A Multi-Agent Large Language Model Framework for Prediction Market Trading and Latency Arbitrage

PolySwarm is a new multi-agent AI framework that uses 50 diverse large language models to trade on prediction markets like Polymarket, combining swarm intelligence with arbitrage strategies. The system outperformed single-model baselines in probability calibration and includes latency arbitrage capabilities to exploit pricing inefficiencies across markets.

• →PolySwarm deploys 50 diverse LLM personas that evaluate binary outcome markets and aggregate predictions through confidence-weighted Bayesian methods.
• →The framework incorporates latency arbitrage modules that exploit stale prices on Polymarket by deriving probabilities from centralized exchange data.
• →Information-theoretic analysis using KL and JS divergence helps detect cross-market inefficiencies and pricing mismatches.
• →Experimental results show swarm aggregation consistently outperforms single-model approaches in probability calibration tasks.
• →The system faces challenges including AI hallucination, computational costs, regulatory exposure, and potential feedback-loop risks.