The updated account clarifies a change in defaults, rather than a complete removal of Anthropic models.
Replit CEO Amjad Masad initially told TechCrunch Disrupt that the company was no longer working with Anthropic. TechCrunch’s updated October 16 report adds a subsequent clarification: customers can still choose Claude and other supported models, while OpenAI is now the preferred partner for Replit Agent.
Masad attributed the preference to performance, economics and the partnership. His claims about application-building performance are Replit’s assessment, rather than an independent comparison. The updated account is more specific than the initial statement and resolves the impression that Claude had disappeared from the platform.
The distinction matters because users can retain a choice even as the default supplier changes. TechCrunch notes that many customers stick with a default. A change in preference can therefore influence usage and spending without an exclusive arrangement.
Why it matters
A default-model change can influence what customers use even when competing models remain available.
For people building on Replit, the practical question is which model runs a task by default and whether a different one can still be selected. A preferred partnership and exclusive availability are different arrangements. The updated account makes that distinction much clearer than the initial onstage remarks.
A default can matter even when alternatives remain available. Someone starting a project may accept the suggested model without comparing providers. That makes the product's routing choices part of the customer's experience of quality, cost and reliability. A model change can therefore affect users who never explicitly decide to switch.
Teams evaluating the change should use representative projects: a new application, a bug in an existing repository and a change that requires tests. Record the completed result, the time spent reviewing it and any additional work needed to make it usable. A benchmark about creating applications from scratch does not settle performance on every maintenance task.
The broader business lesson is that distribution through an application can be as consequential as the model itself. Replit controls the development workflow; the model provider supplies one important component. The preferred supplier can change while the application's relationship with its users continues.
For existing customers, preserving a model choice also provides a way to compare behavior on familiar work. A team can check whether its prompts, tests and review process produce comparable results before standardizing on a default. The clarification is therefore more useful than a simple announcement of a supplier breakup.