Token costs drop out of Replit’s new Free Mode

Replit has launched Free Mode, an agent tier that runs on a single model, GPT-5.6 Luna. In a post on OpenAI’s site, the company pitches it as a way for anyone to turn ideas into working software “without worrying about token costs.” Replit gave the announcement to Fortune first, and Fortune ran it as an exclusive on August 19.
Fortune’s reporting narrows that “anyone.” Free Mode still requires a paid subscription, either Core at $20 a month or Pro at $100 a month, and it is now the default mode for users on both tiers. In Free Mode you can chat, brainstorm or run a simple task before you start spending the larger, limited budget your subscription includes.
Harder tasks get routed elsewhere. When the Replit agent decides a task needs a different or more capable model, it switches automatically “for the duration to get the task done, and then we fall back into Free Mode as soon as possible,” Michele Catasta said, per Fortune. The higher-reasoning model it switches to is GPT-5.6 Sol, and the project context carries over. Fortune also reported that Luna is the only model behind Free Mode, although developers on Replit can normally pick from many models.
The price of Luna is what made Free Mode possible. Catasta credited OpenAI’s 80% cut to Luna’s price on July 30, which took output from $6 to $1.20 per million tokens, the Financial Times reported in August. “A lot of tasks don’t require the frontier-level intelligence. It just requires smaller models that are faster and more affordable,” Catasta said. OpenAI’s Thibault Sottiaux said the cuts came from running the model more efficiently, not from new compute coming online. OpenAI also made the same model available to free ChatGPT users on August 6.
Replit CEO Amjad Masad called model costs “one of the last barriers” to more people building software. He said OpenAI’s price cuts are what let Replit offer Free Mode to millions of users. Fortune places the launch in a wider trend of AI companies lowering prices, as enterprise customers question the return on their AI spending and American model builders compete with cheap Chinese models.
Replit and OpenAI have worked together for years: Replit was an early user of GPT-3, which came out in 2020. Catasta, per Fortune, said: “This is the first of many launches we’re going to be doing together.”
Up next
Claude’s protein binders hit 14 of 15 targets in wet lab, Video
A human expert wrote the prompt, and Claude designed protein binders against 14 out of 15 targets on its own. Between 22% and 35% of them bound successfully, against a typical field rate of 10% to 15%.
Related stories
- OpenAI’s Decisions API exits beta at 400 questions a call
- OpenAI keeps Codex cache forever, API cache gets 12 hours
- Your ChatGPT Plus plan can now pay for other apps’ AI
- Code strings point to a $50 production plan for OpenAI’s API
- GPT-6 caching update cuts cached input costs by up to 90%
- Linear and Ramp users can pay for Claude without an API key
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

