Replit CEO cites cost and performance in Anthropic split

Replit chief executive and co-founder Amjad Masad said on stage at TechCrunch Disrupt that the AI coding platform is no longer working with Anthropic. He said the decision reflected cost, partnership and performance considerations, and argued that OpenAI was a better partner for Replit “for now”.

The statement follows Replit’s August partnership announcement with OpenAI. Replit operates a cloud-based development platform where users can build and deploy applications directly in a browser. Anthropic’s Claude models had run on the platform under the companies’ partnership, which was reported to have helped Replit expand revenue substantially.

Benchmark claims meet conflicting accounts

Masad said Replit uses an internal benchmark to assess measures including performance and also publishes a public benchmark. He said its Sol and Astra products consistently perform better when building vibe-coded applications from scratch, and cited that result as part of the rationale for the change.

However, the status of the relationship is not fully clear. A person familiar with the Anthropic-Replit partnership told TechCrunch that it remained in place. Replit material published in September also cited Anthropic’s Claude Fable 5 and Claude Fable 5.1 as models in Replit Agent’s production environment. TechCrunch sought clarification from Replit after Masad’s comments.

Enterprise becomes a larger revenue source

Masad did not disclose Replit’s annual recurring revenue, but said enterprise contracts represented about 2% of revenue last year and about 15% this year. More than half of the company’s revenue is now B2B, he said, signalling a larger role for commercial customers as the platform grows.

The company’s wider operating model has also stressed AI agents as systems that need controlled environments. In the context of Replit’s AI self-driving company model, Replit’s approach frames agent deployment around orchestration, error recovery, safety, security, sandboxing, networking, identity and authorisation.

What customers should evaluate

Masad compared the requirements for agents to the conditions people need to work effectively: a suitable environment, tools and support. His point was that agents can be deployed safely only after the surrounding habitat is in place.

For businesses adopting AI-assisted development, Replit’s comments underline the need to evaluate providers beyond a headline model choice. Teams should test performance against their own workloads, account for costs and partnership terms, and ensure that security, sandboxing and access controls support production deployment.