Replit debuted Free Mode, a default experience for its $20-per-month Core and $100-per-month Pro subscribers, powered by OpenAI’s GPT-5.6 Luna model. The feature lets users perform unlimited lightweight tasks, including chats, ideation and simple development work, without drawing on their plans’ larger but limited AI-token budgets. When Replit’s agent identifies a task requiring a more sophisticated model, it automatically reroutes the request and returns to Free Mode as soon as possible. Replit President and head of AI Michele Catasta called the approach “radical,” saying it became possible after OpenAI cut Luna’s costs by 80% on July 30. OpenAI’s Thibault Sottiaux said the reduction resulted from operating efficiencies rather than increased compute supply. The launch reflects a broader AI-industry push to improve customer return on investment as enterprise buyers scrutinize spending and competition intensifies, including from cheaper Chinese AI models. Replit and OpenAI said they are working more closely and expect more joint launches. Replit, founded in 2016 by CEO Amjad Masad, was an early user of GPT-3, released in 2020. The companies said they share a goal of making powerful, affordable AI accessible to more people, including users without technical backgrounds who want to engage in vibe coding, or building software through natural-language prompts.