CMS ACO LEAD Model
A policy analysis of the LEAD Model RFA—benchmarking, capitation mechanics, the ACPT savings wedge, and what ACOs should stress-test before applying
Building for LEAD? RevelAi Health helps care organizations stand up the AI-powered care coordination and specialty care arrangements that help you succeed. We partner with ACOs and health systems to bring AI augmented care capabilities to your patients and build reporting infrastructure required to operate in models like ACCESS. Through MSK ACCESS We augment value-based healthcare operations for care organizations with responsible agentic AI superstaffing and automation so you can focus on clinical care, not checkboxes.
→ Book a demo: Here | Contact us by email: Here
Buried on page thirty-one of the 148-page LEAD Model Request for Applications that CMS released today is a sentence worth reading carefully: “LEAD Model ACOs will not have their benchmarks rebased during the Model performance period.”
That’s a ten-year no-rebasing commitment. For anyone who has watched MSSP ACOs lose their efficiency gains to successive benchmark resets—the so-called ratchet effect—this is a meaningful structural concession. But it is not, on its own, sufficient reason to apply. The no-rebasing headline is the front door of a financial architecture that gets considerably more complex once you walk through it, and several of the mechanisms inside deserve additional scrutiny.
The LEAD Model (Long-term Enhanced ACO Design) is CMS Innovation Center’s successor to ACO REACH. Applications opened today, close May 17, 2026, with Performance Year 1 beginning January 2027. It runs through 2036—the longest model test period CMMI has ever proposed.
What follows is my read of the financial architecture as someone who operates in value-based musculoskeletal care and tries to think about ACO economics daily. I’ll focus on the mechanisms that could determine whether LEAD produces margin or loss for participating organizations, and flag the areas where the RFA asks ACOs to commit without providing enough information to model outcomes. If you’re more of a visual learner, I also used notebookLM to create sequential explainers on different aspects of the model. A 1 hour crash course with captions pulling directly from the RFA, CARA documents, webinar transcript and more.
