Replit drops Anthropic as preferred partner and makes OpenAI its default

Replit says it has moved on from Anthropic

Replit, the browser-based AI coding platform, is no longer working with Anthropic, according to CEO and co-founder Amjad Masad, who made the announcement on stage at TechCrunch Disrupt. TechCrunch reports that the decision reflects a combination of cost, partnership quality and model performance, with OpenAI now the preferred backer of Replit Agent.

Masad was blunt about the performance side of the reasoning. "I think ChatGPT is better now," he said, explaining that Replit evaluates models using an internal benchmark alongside a published public one. According to Masad, models he named Sol and Astra consistently outperform alternatives when building "vibe-coded applications" from scratch. "It was a decision based on both cost, partnership, and performance," he said.

A quiet shift after the OpenAI deal

Replit announced a partnership with OpenAI in August, but until now there was no public indication that its relationship with Anthropic had ended. Before the OpenAI deal, Replit's platform ran on Anthropic's Claude models, an arrangement TechCrunch says helped the company grow its revenue significantly. Masad noted on stage that Replit had been, until recently, one of Anthropic's top five customers.

Anthropic models remain, but not by default

The break is less total than the on-stage statement suggested. After the event, Masad clarified that customers can still choose Anthropic models on the platform, but OpenAI's ChatGPT is now the default option. That detail matters commercially: as TechCrunch points out, most customers stick with the preselected model rather than picking one themselves, so defaulting to OpenAI could sharply reduce Replit's spending with Anthropic even if Claude technically remains available.

"Replit remains model agnostic," Masad told TechCrunch. "Our users can choose Anthropic's Claude or other supported models… OpenAI is now our preferred partner for powering Replit Agent, based on performance, economics, and the strength of our partnership."

Insiders dispute the breakup

Not everyone agrees the partnership is over. TechCrunch reports that Masad's on-stage comments were rejected by people familiar with the Anthropic–Replit relationship, one of whom, speaking anonymously, said the partnership was still in place. Public evidence also points the other way: a Replit blog post from late September cited Anthropic's Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent's production environment, and a Replit product employee recently promoted Anthropic's flagship Opus 5.5 on X.

Enterprise now drives a growing share of revenue

Speaking at the same event, Masad described how Replit's business mix is changing. He declined to share ARR figures but said enterprise contracts went from roughly 2% of revenue last year to about 15% this year, with more than half of revenue now coming from B2B. Backstage, he told TechCrunch he expects enterprise and government customers to account for more than half of revenue in 2027.

He also touched on AI safety, arguing that agents need a suitable "habitat" to do good work, including orchestration, error recovery, sandboxing, networking, and identity and authorization controls. "Once you figure all of that out, then you can deploy AI safely," he said.

Why it matters

Replit was reportedly one of Anthropic's biggest customers, and its move to make OpenAI the default for Replit Agent shows how quickly loyalty in the AI tooling market can shift when benchmarks, pricing and partnership terms change. For Anthropic, losing default placement on a popular coding platform is a meaningful revenue and distribution setback, even if its models technically remain available. For the broader market, the episode is a reminder that partnership announcements often reveal less than they seem to: Replit's August OpenAI deal implied nothing about Anthropic's status, and even now the two sides disagree about whether the relationship has actually ended. As AI coding platforms consolidate around fewer providers, default model placement — not mere availability — is becoming the real battleground.