Replit launches Free Mode on OpenAI's cut-price Luna model

Replit's launch of Free Mode marks a strategic shift in how coding platforms deliver AI value, underpinned by OpenAI's dramatic 80% cost reduction on Luna announced July 30. Rather than charging users per AI inference, Replit now includes a capable—if less powerful—model by default in all paid tiers, reserving premium models for tasks that genuinely demand them. This tiered approach solves a key friction point: users on subscription plans often hesitate to use AI features for fear of rapidly exhausting their token limits.

The partnership reflects deeper industry pressures. As enterprise customers scrutinize return on investment from AI spending, and as competition from low-cost Chinese AI models accelerates, American AI vendors are competing on affordability and breadth rather than raw capability alone. OpenAI's efficiency gains on Luna—not new hardware, according to Sottiaux—suggest the company has found ways to run inference more cheaply, making it viable for platforms like Replit to offer AI-powered features at scale without unsustainable margins.

Both companies frame this as the opening salvo in a broader collaboration. Catasta and Sottiaux emphasize their shared mission to democratize AI access, particularly for non-technical users. Sam Altman's public statement envisions a world where AI is as commodified as utilities, removing barriers to entrepreneurship. Replit's stance—calling the move 'radical'—suggests this level of AI inclusion in a base-tier product is still unusual in the industry, even as cost pressure mounts.