Replit CEO says the platform is no longer working with Anthropic

Amjad Masad, co-founder and CEO of the AI software development platform Replit, said on Thursday at the TechCrunch Disrupt technology conference that his company is no longer working with the AI lab Anthropic. According to TechCrunch, Replit has become popular among startup founders and developers. The move matters because the platform had been running on Anthropic's Claude models.

Reasons for the decision

Masad said several factors were behind the decision, including cost. He said OpenAI, the maker of ChatGPT, now offers the better option. Replit uses an internal benchmark to assess metrics such as performance, and also publishes a public benchmark. According to Masad, the models Sol and Astra consistently do better at building so-called vibe-coded applications from scratch. He described the decision as based on cost, partnership and performance.

Replit announced a partnership with OpenAI in August, but nothing had indicated that its relationship with Anthropic had ended. Replit sells a cloud-based development platform where users can build and deploy apps directly in a browser. The Anthropic partnership was said to have helped the company grow its revenue significantly.

Claim disputed

Masad's statement drew pushback from people familiar with the relationship between the two companies. One of them, who asked not to be named, said the partnership remained in place. A September blog post also cites Anthropic's Claude Fable 5 and Claude Fable 5.1 as models running in the production environment of Replit Agent. TechCrunch has asked Replit for clarification and will update the story if it responds.

Growth and safety

Masad said enterprise customers are taking a larger share of revenue. Enterprise contracts made up about 2% of revenue last year, he said, and about 15% this year, while more than 50% of revenue is business-to-business (B2B). He did not share annual recurring revenue figures.

He also stressed the need for AI safety measures. Agents, he said, need a "habitat" with good orchestration, error recovery, security, sandboxing, networking, and identity and authorization. Once that is in place, AI can be deployed safely.