Replit, the AI coding platform popular among startup founders and developers, has ended its collaboration with Anthropic, as confirmed by CEO and co-founder Amjad Masad during a presentation at TechCrunch Disrupt.
Masad explained this decision stemmed from a variety of factors, including financial considerations. Although Replit announced a partnership with OpenAI in August, there was no previous indication of a termination of the relationship with Anthropic.
“I believe ChatGPT has improved,” Masad stated, noting that the company employs an internal benchmark to assess metrics such as performance. “We also publish a public benchmark, and consistently, Sol and Astra excel in creating vibe-coded applications from scratch. Our choice was influenced by costs, partnership dynamics, and performance; thus, we concluded that this was a more suitable collaboration for now.”
However, it appears that Replit hasn’t completely severed ties with Anthropic. The platform, which allows users to develop and deploy applications directly in their browser, will still provide access to Anthropic models, although OpenAI’s ChatGPT has now taken precedence as the default option. This change is likely to decrease Replit’s expenditure with Anthropic, as most users tend to stick with the default model.
Before collaborating with OpenAI, Replit utilized Anthropic’s Claude models for its operations, a partnership that significantly contributed to the company’s revenue growth. Masad mentioned during his presentation that Replit was, until recently, among Anthropic’s top five clients.
“Replit continues to be model agnostic,” Masad clarified to TechCrunch over the weekend. “Our users can select from Anthropic’s Claude or other supported models… OpenAI is currently our preferred partner for powering Replit Agent, a decision based on performance, cost-effectiveness, and the strength of our alliance.”
Despite Masad’s statements, some insiders familiar with the arrangement between Replit and Anthropic assert that the partnership remains intact.
Evidence from the company’s website, blog entries, and social media indicates that Replit was still endorsing Anthropic products recently. For example, a blog post from late September mentioned Claude Fable 5 and Claude Fable 5.1 as models deployed within Replit Agent’s production environment. Simultaneously, a Replit product employee shared on social media about utilizing Anthropic’s premier model, Opus 5.5, highlighting its capabilities.
While on stage at Disrupt, Masad shared insights regarding the company’s growth trajectory, mentioning that enterprise clients are starting to make up a larger share of revenue.
“We’re not revealing ARR figures at the moment, but in terms of growth, enterprise revenue was about 2% last year,” Masad indicated. “This year, I estimate that enterprise contracts account for approximately 15% of our revenue, and B2B makes up over 50% of our earnings.”
Backstage, Masad told TechCrunch that he anticipates that more than half of Replit’s revenue will derive from enterprise and government clients by 2027.
In addition, Masad stressed the importance of robust AI safety measures.
“Just as humans require an appropriate environment to work—like a comfortable office, good food, and efficient technology—agents also need a suitable habitat to function effectively,” he explained. “This habitat involves good orchestration, effective error recovery, solid safety and security protocols, as well as adequate networking and authorization. Once these elements are in place, AI can be deployed safely.”
This article was initially published on October 16 at 2:22 pm PT and has been updated for additional context regarding Replit’s partnerships with Anthropic and OpenAI.
