Replit Makes OpenAI the Default but Keeps Anthropic Models Available

Replit has made OpenAI its preferred provider for Replit Agent, but it has not removed Anthropic from the platform. CEO Amjad Masad said the decision reflected cost, partnership quality and performance on internal and public benchmarks. He later clarified that customers can still choose Anthropic's Claude and other supported models.

That distinction matters because defaults shape usage. Most customers do not change a preselected model, so the move could redirect a meaningful share of Replit's inference spending toward OpenAI even while Anthropic remains available.

Economics meet product performance

Replit previously relied heavily on Claude, and Masad said the company had been one of Anthropic's five largest customers. The new default therefore has commercial significance for both model providers. For Replit, lower model cost or better application-building performance can improve gross margin and user experience at the same time.

The company says it uses benchmarks that test whether models can build applications from scratch. Masad cited stronger results from OpenAI models, but benchmark results do not capture every enterprise requirement. Reliability, security, latency and the ability to work inside existing codebases also affect customer choice.

Replit's own August product announcement showed that OpenAI already powered its Free Mode. The broader preferred-partner status extends that relationship, while Replit remains model agnostic rather than committing exclusively to one vendor.

Enterprise ambitions raise the stakes

Masad said enterprise represented about 15% of revenue this year, up from roughly 2% last year, and that business customers accounted for more than half of revenue. He expects enterprise and government to exceed half of revenue in 2027. Those figures were management comments, not audited segment disclosures, but they show why provider economics and security matter more as larger customers adopt the product.

The change is a useful signal of competition in foundation models: providers increasingly win distribution through total operating value, not raw benchmark scores alone. Replit can also use its multi-model option as leverage in future negotiations.

The key evidence will be whether model costs fall, enterprise adoption continues and customers actively choose alternatives to the default. Until Replit discloses more detail, the immediate conclusion is narrower: OpenAI gained preferred distribution, while Anthropic retained access to Replit users.