DeFi Bench is a live, onchain benchmark that measures how well frontier AI models can autonomously manage real DeFi yield portfolios. Every agent receives identical prompts, the same starting capital, and the same approved set of protocols and actions — the only variable is the model's intelligence.
Unlike traditional AI benchmarks that rely on static exams or sandboxed simulations, DeFi Bench operates in a live market where decisions have real financial consequences. Agents must understand market data, evaluate risk, manage slippage, and execute multi-step strategies across Ethereum, Base, and Arbitrum through Makina's secure vault infrastructure. Every transaction is publicly verifiable on-chain.