Replit, the browser-based platform for building and deploying apps, has moved away from Anthropic as its default model provider, according to CEO Amjad Masad. He made the comments onstage at TechCrunch Disrupt, as reported by TechCrunch.
What Masad said
Masad said the decision rested on several reasons, cost among them. He argued that ChatGPT is now better for Replit's purposes and pointed to an internal benchmark and a public one. According to him, two models he named Sol and Astra consistently do better at building applications from scratch. He summed it up as a call based on cost, partnership and performance.
After the session he clarified that Anthropic is not fully gone. Customers can still pick Anthropic's models on the platform, but OpenAI's ChatGPT is now the default. He also described OpenAI as the preferred partner for powering Replit Agent, and said Replit stays model agnostic.
Why the default matters
TechCrunch notes that most customers keep the default instead of choosing a model themselves. That means the switch could cut Replit's spend with Anthropic substantially. Masad said Replit had until recently been one of Anthropic's top five customers. Before the OpenAI deal, announced in August, Replit relied on Anthropic's Claude models to run the platform.
What is disputed
The account is not uncontested. One person familiar with the Anthropic and Replit relationship, who asked not to be named, told TechCrunch the partnership remains in place. This is unconfirmed by either company on the record in the source. TechCrunch also found that a late-September Replit blog post named two Claude models as running in Replit Agent's production environment, and that a Replit product employee posted on X about using an Anthropic flagship model around the same time.
Other points from the session
Masad said enterprise was about 2% of Replit's revenue last year and is about 15% this year, with more than 50% of revenue coming from B2B. He told TechCrunch he expects more than half of 2027 revenue to come from enterprise and government customers. He also said agents need a controlled environment with orchestration, error recovery, sandboxing, networking, identity and authorization before they can be deployed safely.