Meerada Builds a Live Market for AI Models

Field notes

The exchange organizes models around verified task outcomes rather than token pricing alone, using a stated CPAT measure and a grading battery that includes hidden unit tests, real SQL, exact answers, and faithful summaries. Its public board exposes confidence intervals, sample counts, trends, and update status, while the browser cockpit is intended to run multiple models on multiple tasks in parallel. Coverage depends partly on added free-tier keys and sponsored runs for paid frontier models.