Replit makes OpenAI the default for Replit Agent: how to keep building with Claude, and what changes for your apps

Replit has moved its coding agent’s default from Anthropic to OpenAI. On stage at TechCrunch Disrupt, CEO Amjad Masad said Replit is “no longer working with Anthropic”, then clarified afterwards that customers can still use Anthropic models and that OpenAI’s models are now the default. (TechCrunch)

His written statement to TechCrunch is the version builders should work from: “Replit remains model agnostic. Our users can choose Anthropic’s Claude or other supported models… OpenAI is now our preferred partner for powering Replit Agent, based on performance, economics, and the strength of our partnership.” TechCrunch also reports that people familiar with the Anthropic relationship say the partnership is still in place, so treat the business side as disputed. The product side is what you can check.

Key facts:

• Why: Masad cited cost, partnership and performance, and said OpenAI’s Sol and Astra models “consistently” do better on Replit’s internal and public benchmarks for building vibe-coded apps from scratch.
• Claude is still there: the Model selector docs list Claude Sonnet 5 and Sonnet 4.6 in Power Mode, and Claude Fable 5.1, Opus 5.5, Opus 5, Opus 5 Fast and Opus 4.8 in Max Mode.
• Default matters: TechCrunch notes most customers stick with the default rather than choosing a model, which is why the switch is a big shift in Replit’s spend even with Claude still available.
• Background: Replit’s August Free Mode launch already ran on OpenAI’s GPT-5.6 Luna, announced jointly by Replit and OpenAI.

How to keep using Claude in Replit Agent

Per Replit’s docs, only builders on Core and Pro can pick a model, and only in Power or Max mode:

1. In the prompt box, open Agent Mode and choose Power or Max.
2. Find the Primary model section and pick a Claude model from the menu.
3. The choice applies to your next message, and you can change it at any time.

Some limits apply. Free Mode always uses Replit’s automatic routing and has no model menu. Auto in Power or Max also lets Replit choose the model, so if you want Claude every time, pick it by name. The Fast variants, such as Opus 5 Fast, need Pro or Enterprise. Replit says model selection is still rolling out: if you don’t see Primary model, the selector isn’t on your account yet. Your own menu is the source of truth for what you can pick.

Separately, Effort (Low through Max) controls how much reasoning the chosen model uses. Replit warns that higher Effort can cost up to about 2x on the hardest tasks.

Enterprise: who controls the list

On Enterprise, account admins set provider policies under Settings > Advanced > Agent modes and model providers: Enable all, Enable selected per Workspace, or Disabled. Replit requires at least one model each from OpenAI, Anthropic and Google to stay enabled, so an admin can’t remove Claude completely. Enterprise Auto mode can’t pick a model that policy has turned off. (Replit docs: Agent modes and models)

Your app’s model is a separate setting

The agent that writes your code and the model your finished app calls are configured separately. If you ask Agent to build an app with Anthropic, Replit AI Integrations still offers Claude models, including claude-opus-5, claude-sonnet-5 and claude-haiku-4-5, billed to your Replit account. Look for “Anthropic (Replit managed)” in Agent’s reply, or bring your own key to be billed by Anthropic directly. Integrations require a paid plan and are off by default for Pro and Enterprise organizations until an admin turns them on.

To compare the two families yourself, Replit’s beta Compare models option runs one prompt on up to four models side by side, with each run billed separately. For the models involved, see our coverage of GPT-6 Sol and Luna’s launch, GPT-6.1 Sol versus Astra on cost and Claude Opus 5.