The partnership between Replit and Anthropic suddenly became hard to describe on October 16, 2026. TechCrunch reported that day that Replit CEO Amjad Masad, asked onstage at TechCrunch Disrupt whether the company's new OpenAI partnership meant it was no longer working with Anthropic, answered in the affirmative. Yet the same report cited a person familiar with the partnership who said it was still in place. With the CEO's public statement colliding with an insider's denial, the relationship has shifted at least in how it is described.
The CEO's stated reasons: cost, partnership and performance
Masad attributed the decision to three things: cost, partnership and performance. He said Replit uses an internal benchmark called VBench as well as a public benchmark, and that on those measures OpenAI's Sol and Astra consistently perform better at building vibe-coded applications from scratch. Replit had announced its OpenAI partnership back in August, with no indication at the time that the Anthropic relationship was ending alongside it; Masad also noted that Replit had been one of Anthropic's largest customers. Replit's platform lets users build and deploy apps directly in the browser, and Claude models previously powered its agent products, a partnership credited with helping drive significant revenue growth.
The contradictory evidence is only a few weeks old
The insider's denial is not baseless. As recently as late September, a Replit blog post cited Claude Fable 5 and Fable 5.1 running in its agent's production environment, and a product employee was publicly discussing work with Opus 5.5 around the same time. If the partnership had truly ended outright, that material is hard to explain; if core workloads merely moved while integrations remain, the onstage wording sounds overly absolute. TechCrunch has asked Replit for clarification and had not received it when the story ran.
The deeper story is how the relationship between coding platforms and model labs is changing. Platforms know better than anyone which model is most economical on their real workloads, and they are increasingly unwilling to bet the company on a single supplier, while OpenAI and Anthropic push into building and hosting territory that belongs to the platforms, making suppliers into competitors. Our coverage of Claude Code Review's official launch already showed labs pulling more of the coding chain in-house. For developers the takeaway is practical: check that a platform supports switching between models, and avoid binding your workflow deeply to one lab's proprietary capabilities, because a change in wording between suppliers can quickly become your migration cost.
