Replit CEO says company has stopped working with Anthropic

What is new

Replit CEO and co-founder Amjad Masad said at TechCrunch Disrupt in San Francisco on Thursday that Replit is no longer working with Anthropic, the AI lab behind the Claude models. Under that partnership, Replit used Claude models to run its platform, according to TechCrunch.

Masad said the decision was made for a number of reasons, including cost. He described it as a decision based on cost, partnership and performance, and said the company felt OpenAI was “a better partner for now.” He said Replit uses an internal benchmark and also publishes a public one. He said models called Sol and Astra perform better at building vibe-coded applications from scratch. The sources do not give benchmark results.

Replit announced a partnership with OpenAI in August. Replit sells a cloud-based software development platform where users can develop and deploy apps directly from their browser.

Separately, Masad described his vision of a “self-driving company,” where agents handle much of the routine work at a company and people are valued more for their judgment and taste. He used Replit’s own support process as an example. When support tickets come in, an agent first examines them by looking at GitHub issues, the codebase and Linear tickets. By the time a person sees the request, they have the extra information needed to decide how to proceed. When that support person submits a patch to apply a fix, it takes one click to deploy another agent. Masad said, “You need human judgment; you need taste.”

Masad also gave figures on the company’s growth. He said enterprise was about 2% of revenue last year. This year, he said, about 15% of revenue comes from enterprise contracts, and more than 50% of revenue is B2B. He said Replit is not sharing ARR numbers yet.

On safety, Masad said agents need a “habitat” to do good work. He listed good orchestration, error recovery, safety and security, sandboxing, networking, and identity and authorization as parts of that habitat. He said that once these are worked out, a company can deploy AI safely.

Limits, privacy and academic integrity

The claim that Replit has stopped working with Anthropic is disputed in the same TechCrunch report. One person familiar with the partnership, who asked not to be named, said the partnership was still in place.

TechCrunch points to public evidence that Replit was using Anthropic products as recently as a few weeks before the talk. A blog post from late September cites Claude Fable 5 and Claude Fable 5.1 as models running in the production environment of Replit Agent. Around the same time, a Replit product employee posted on X about using Anthropic’s flagship model, Opus 5.5, and promoted its capabilities.

TechCrunch said it reached out to Replit for clarification and would update the story if it heard back. The article was later updated to add more context on Replit’s use of Anthropic models.

The performance comparison comes from Masad’s own account of Replit’s benchmarks, and the sources do not include the results. The revenue percentages were also given in an onstage talk, and ARR was not disclosed.