Dolores Research launches WAGMI Bench, an open-source benchmark for AI trading agents on Virtuals Protocol

Dolores Research just dropped an open-source framework designed to do for AI trading agents what SWE-bench did for coding agents: give everyone a standardized, reproducible way to measure whether these things actually work. The project, called WAGMI Bench, evaluates AI trading agents across 13 historical periods in the BTC perpetuals market, covering everything from the COVID crash to the window around ETH approvals. Each agent faces 3,150 decision points under consistent rules. No cherry-picking favorable time windows, no conveniently omitting the drawdowns. What WAGMI Bench actually measures The benchmark’s core philosophy is a deliberate departure from the standard crypto trading flex. Instead of leading with profit-and-loss numbers, WAGMI Bench evaluates agents on survival and engagement metrics. The initial results from the Classic 13 study (v1) tested five models plus one mechanical baseline. Dolores Research hasn’t positioned any single model as a winner. The emphasis is on the framework itself, not on crowning a champion. The project is released under an Apache-2.0 license, meaning anyone can fork it, extend it, or run their own agents through the gauntlet. The inspiration ...