TechCrunch reported that Replit CEO Amjad Masad said onstage at Disrupt on Thursday that the company no longer works with Anthropic. Replit sells a browser-based tool for building apps, and had used Anthropic's Claude models to run it. Masad cited cost, partnership and performance. He said Replit keeps an internal benchmark and also publishes a public one, and that the models he named, Sol and Astra, consistently do better at building vibe-coded apps from scratch. Both benchmarks are Replit's, so this is the CEO's claim.
An unnamed person familiar with the Anthropic and Replit ties disputes this, TechCrunch said, and says the deal continues. TechCrunch also pointed to a September blog post that cites two Claude models as running in the production environment of Replit Agent. The sources do not say when any change took effect. They give no contract terms and no comment from Anthropic.
The model under an AI app builder is a part the buyer cannot see and the vendor may swap. A platform like Replit passes work to a model maker's system, so that model shapes both its cost and its output quality. A from-scratch benchmark scores how well a model builds an app from a blank start. Buyers can ask which models run the product, whether they can change without notice, and how changes are announced.