Replit Switches Its Default AI to OpenAI, but Claude Remains an Option

Replit has changed the default model provider behind Replit Agent, its app-building assistant, to OpenAI. According to TechCrunch, Replit CEO Amjad Masad said the company chose OpenAI as its preferred partner. He added that Anthropic’s Claude models remain available for users to select, within the range that Replit supports.

Masad pointed to performance, cost, and the partnership itself as the reasons for the switch. Users who do not explicitly specify a model will now be served OpenAI models. For those who had been working with Claude, the ability to manually choose a model for a given task has become more important.

Can you still use Claude?

Yes. Replit has shipped products in partnership with Anthropic for some time. Agent v2, released in 2025, ran on Claude 3.7 Sonnet. In June 2026, the company introduced a feature that lets users hand off app-building work started in Claude to Replit. In a technical post that September, Replit said it was using both Claude and OpenAI models to run Replit Agent.

Accounts of the relationship between the two companies have been inconsistent. Speaking on stage at an event, Masad said Replit was no longer working with Anthropic. He later clarified that Claude remains usable. An anonymous source cited by TechCrunch also maintained that the two companies are still collaborating.

What changes about model selection?

Replit already has a feature that picks the right model for a given task on its own. Intelligent Model Routing, announced on August 26, weighs output quality, processing speed, and cost together when deciding which model to use. It is available to all users, sparing them the effort of comparing models task by task.

Whether you can choose a model yourself depends on your plan and workspace settings. Core and Pro users can specify a model directly. Enterprise administrators can decide which models are permitted in a workspace. For users on a company account, Replit automatically selects a task-appropriate model from the list the administrator has approved.

Is the cost really 65% lower?

Replit says it ran its own tests on the automatic model selection feature released in August and achieved quality equivalent to the previous Max Mode at 65% lower cost. That 65% figure comes from testing of the automatic selection feature.

Free Mode, a separate feature, also runs on OpenAI models. Replit says Free Mode lets Core subscribers build up to 30 times more than before, with up to 30 hours of chat per month. The Core price listed in that announcement is $20 per month.

Users who want to learn the ropes in Korean can consult Replit’s official Korean-language Agent documentation. In the end, only one thing has really changed: the default model is now OpenAI, but the Claude option remains. What to watch next is how the new default and the automatic selection feature interact in real work, and what difference that makes to the cost and quality of building apps.