Replit, the browser-based platform where people build and deploy software with an AI coding agent, has made OpenAI the default model behind that agent, and its chief executive first described the change as a break with Anthropic. Amjad Masad told a TechCrunch Disrupt audience on Thursday that Replit had stopped working with Anthropic. Afterward he described something much narrower. TechCrunch, which reported both versions, updated its article with more context on Replit’s partnerships and changed the title.
The distance between the two accounts is the story. On stage, Masad gave cost, performance and partnership as the reasons and said, “I think ChatGPT is better now.” In a later statement to TechCrunch he said “Replit remains model agnostic,” which means the platform does not tie itself to a single AI supplier, and that users can still pick Anthropic’s Claude or other supported models. OpenAI, he said, is “our preferred partner for powering Replit Agent.”
A default is a commercial decision, whatever the menu says. TechCrunch points out that most customers stay with the default rather than choosing a model themselves, so Replit’s spend with Anthropic could fall sharply even though Claude remains selectable. Masad also said on stage that, until recently, Anthropic counted Replit among its five biggest customers. Losing the default slot at a customer of that rank is a revenue event for Anthropic, not a product preference. Neither the size of that spend nor the size of the shift has been published.
Replit’s history with Anthropic makes the move sharper. Replit announced its OpenAI deal in August. Until then it ran its platform on Claude models, a relationship TechCrunch describes as having contributed substantially to the company’s revenue growth.
Then there is the dispute over what is actually true. People familiar with the Anthropic and Replit relationship rejected Masad’s on-stage account, and one of them, who asked not to be named, said the partnership was still in place. TechCrunch also found Replit promoting Anthropic only weeks earlier. A late-September blog post named Claude Fable 5 and Claude Fable 5.1 among the models in Replit Agent’s production environment, and around the same time a Replit product employee posted on X about using Opus 5.5. Those posts show where things stood in September. They do not show what the default is now, and both can hold at once if the switch was recent.
The performance claim belongs to Replit alone. Masad cited an internal benchmark and a public one that Replit publishes, and said the models he called Sol and Astra do better at building applications from scratch. TechCrunch names no independent evaluation. A company that lists cost among its reasons and then justifies performance with its own tests has given reasons that outsiders cannot check.
The business backdrop explains why Replit would want both a new preferred partner and the freedom to say it is not locked in. Masad said enterprise contracts were about 2 percent of Replit’s revenue last year and are about 15 percent this year, and that more than half of revenue is now business-to-business. Backstage, he told TechCrunch he expects government and enterprise buyers to supply over half of Replit’s 2027 revenue. He declined to share annual recurring revenue. Large buyers tend to want leverage over suppliers, and “model agnostic” is the posture that keeps that leverage.
For anyone building on Replit, the practical question is which model your agent now runs on, because a changed default alters cost and results without anyone clicking anything. For Anthropic, the question is how many other big customers are one default setting away from the same move.
Reported by TechCrunch on 17 October 2026
