What happened
On August 19, Replit announced Free Mode, now the default for its $20/month Core and $100/month Pro plans. Powered exclusively by OpenAI's GPT-5.6 Luna, the feature lets paid subscribers run unlimited lightweight chat, ideation, and routine tasks without burning through their AI credit allowances. Replit president Michele Catasta called it "radical" and credited OpenAI's July 30 price cut—Luna dropped 80%, from $1.00 to $0.20 per million input tokens—for making the economics work. When a task requires more horsepower, Replit's agent routes to Sol, then returns to Free Mode.
The numbers
Replit says Core users now get 30x more usage—scaling from roughly 1 hour to 30 hours of Luna-powered chat per month. Luna now costs $0.20 input and $1.20 output per million tokens, down from the July 9 launch price of $1.00 and $6.00. Sam Altman and Catasta both indicated this is "the first of many launches" between the companies, with Altman predicting a "renaissance-level entrepreneurial boom" as cheap, capable models become infrastructure. OpenAI's Thibault Sottiaux said the cut came from efficiency gains, not more compute supply.
Why it matters
Replit is arbitraging OpenAI's pricing to create user value, but the value is contingent, not owned. When Luna's price changes—or its intelligence ceiling shifts—Replit's Free Mode math breaks, and your "unlimited" tier becomes metered overnight. The product looks like a win, but the stack dependency just deepened: you now rely on Replit relying on OpenAI's model family, pricing roadmap, and API stability. Cheaper intelligence is real, but renting it at scale doesn't make it yours—it makes the landlord more powerful.