Replit's chief executive, Amjad Masad, told a packed room at TechCrunch Disrupt on Thursday that his company has walked away from its partnership with Anthropic, pointing to cost, performance and "partnership" as the reasons. "I think ChatGPT is better now," Masad said from the stage, according to TechCrunch, adding that Replit's internal and public benchmarks show OpenAI's Sol and Astra models "perform better on building vibe-coded applications from scratch."
The claim landed as news because Replit, an AI coding platform popular with startup founders, had built much of its product and its public reputation on Anthropic's Claude models. Masad described OpenAI, which Replit announced a partnership with in August, as "a better partner for now."
A Claim at Odds With the Evidence
TechCrunch's report notes real friction with Masad's version of events. A person familiar with the Replit-Anthropic relationship, who was not authorized to speak publicly, told the outlet the partnership was still in place as of this week. Replit's own documentation and blog posts, including one published in late September, listed Anthropic's Claude Fable 5 and Fable 5.1 as models running inside Replit Agent's production environment. A Replit product employee also posted on social media around the same time about using Anthropic's flagship Opus 5.5 model. TechCrunch said it had asked Replit for clarification and would update its reporting if the company responded.
The contradiction leaves the actual state of the relationship unclear: either Replit has quietly begun unwinding a dependency it has relied on for years, or Masad's on-stage remarks overstated a shift that is still in progress, or limited to certain workloads. Either reading says something about how AI application companies talk about the model providers they depend on, and how quickly that framing can change in public.
Betting Big on Enterprise
Beyond the Anthropic dispute, Masad used the appearance to describe a shift in Replit's customer base. Enterprise contracts made up roughly 2% of Replit's revenue last year; this year, he said, they account for about 15%, with more than half of all revenue now coming from business-to-business sales. He did not disclose a specific annual recurring revenue figure.
Masad also addressed AI safety, describing agents as needing a "habitat" — his term for the orchestration, error recovery, sandboxing, networking and identity controls that let autonomous coding tools operate without causing harm. He argued companies can deploy AI agents safely once those safeguards are in place, a claim that comes as regulators scrutinize AI firms' safety practices more broadly.
Replit has not issued a public statement reconciling Masad's remarks with its own technical documentation. Until it does, the specifics of which models power which parts of its platform, and how firmly Replit has actually moved away from Anthropic, remain an open question.