Replit CEO and co-founder Amjad Masad said the AI coding platform is no longer working with Anthropic. He made the comment onstage at TechCrunch Disrupt, describing a decision shaped by cost, partnership considerations and model performance. The status of the relationship is not clear, however: a person familiar with the partnership said it was still in place.
Masad points to cost and model performance
Masad said Replit chose a different partner for several reasons. He said the company uses an internal benchmark to assess measures such as performance, and also publishes a public benchmark. He described the company’s assessment as favoring ChatGPT.
“I think ChatGPT is better now,” Masad said. He added that “Sol and Astra perform better on building vibe-coded applications from scratch.” He connected that assessment with cost and partnership considerations, saying Replit felt the other partner was a better fit “for now.”
The comments offer a glimpse into how Replit weighs AI model providers. The decision, as Masad described it, was not based on performance alone. Cost and the broader partnership also mattered, though he did not provide further details about how those considerations compared.
The partnership’s status remains disputed
Replit announced a partnership with OpenAI in August. The report notes that the announcement did not indicate Replit had stopped working with Anthropic. That leaves open whether the newer partnership replaced the earlier one or whether the companies’ arrangements overlap.
There is also evidence of Anthropic models being used recently. A Replit blog post in September cited Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent’s production environment. Separately, a person familiar with the Anthropic and Replit partnership, who asked not to be named, said the partnership was still in place.
Those accounts do not resolve the disagreement. Masad’s onstage comments describe Replit as no longer working with Anthropic, while the unnamed source said the partnership remained active. TechCrunch said it had reached back out to Replit for clarification and would update its story if it heard back. The report did not include a response from the company.
Replit’s business is shifting toward enterprise
Replit provides a cloud-based software development platform that lets users build and deploy apps from a browser. The report says its Anthropic partnership involved using Claude models to run the platform and was said to have helped Replit grow revenue significantly.
Masad also discussed the changing mix of Replit’s business. He said enterprise accounted for 2% of revenue last year and that enterprise contracts represented 15% this year. He added that more than 50% of revenue was B2B. He did not share annual recurring revenue figures, saying, “We’re not sharing ARR numbers just yet.”
The figures distinguish enterprise contracts from the broader business-to-business share Masad described. Together, they indicate that enterprise is becoming a larger part of revenue, while B2B already accounts for more than half. Masad did not provide additional detail about what drove those changes.
Masad says safe AI needs the right infrastructure
Masad linked responsible AI deployment to the environment in which agents operate. He compared the needs of AI agents with the conditions people need to work, then described infrastructure and safeguards that he said support agents doing good work.
His list included orchestration, error recovery, safety and security, sandboxing, networking, identity and authorization. In his framing, deploying AI safely depends on the surrounding systems as well as the models themselves. That emphasis connects the partnership discussion to the practical demands of running AI coding tools in production.
For now, the central question remains whether Replit’s relationship with Anthropic has ended. Masad’s comments suggest a change in partner, but the September model references and the account from a person familiar with the partnership point to a more complicated picture. Replit’s clarification could establish how its model partnerships and production setup have changed.