TL;DR
Replit is making OpenAI its preferred partner for Replit Agent, while Anthropic’s Claude remains available through the model selector. The shift follows CEO Amjad Masad’s comments at TechCrunch Disrupt and a subsequent clarification. Replit cites performance, economics, and partnership terms; its comparative benchmark claims are company-reported.
What Masad said onstage
Speaking at TechCrunch Disrupt, Replit CEO and co-founder Amjad Masad initially described the Anthropic relationship as ending, according to TechCrunch. His subsequent clarification described a change in preferred partner and default model, with Claude still available. Replit's August announcement introduced an OpenAI-powered Free Mode.
TechCrunch reports that Masad cited cost, partnership terms, and Replit's own model evaluations as reasons for preferring OpenAI. His comparative performance claims describe Replit's assessment of its application-building workflow, rather than a general ranking of all models.
The other version
TechCrunch also reported that a person familiar with the Anthropic partnership said it remained in place. Replit's September engineering post describes Claude and OpenAI models operating in its agent system, consistent with a platform that supports multiple providers.
In his clarification to TechCrunch, Masad said OpenAI had become the preferred partner for Replit Agent while users could still choose Claude. Replit's model-selector documentation lists Claude options for Core and Pro users in Power or Max Mode. Availability depends on plan, mode, rollout, organization settings, and authorization. Free Mode uses automatic routing and does not offer manual model selection.
What the default could change
A default change can redirect usage toward a different provider. D-CSIL analysis: this could reduce Anthropic’s share of Replit’s model spending, but no public usage or spending breakdown establishes the size of that shift. TechCrunch reported Masad's figures for enterprise contracts and B2B revenue; those remain company statements rather than independently audited results presented here.
The comparative benchmark claims are company-reported and not independently verified for this briefing. Replit's engineering blog supplies its harness design, evaluation settings, costs, and scores. Those results should be assessed in that context, rather than treated as proof of a universal advantage.