Replit says it has dropped Anthropic; insiders dispute that

Amjad Masad, chief executive and co-founder of Replit — an AI-assisted coding platform widely used by startup founders and developers — said Thursday onstage at TechCrunch Disrupt that the company is no longer working with Anthropic. Masad attributed the decision to a number of factors, including cost. Replit announced a partnership with OpenAI in August, but until Thursday there had been no indication that its Anthropic partnership had ended.

Replit offers a cloud-hosted software development platform that lets users build and ship apps directly from the browser. Under its partnership with Anthropic, the company ran that platform on the AI lab’s Claude models — a relationship Masad said had previously helped drive significant revenue growth for Replit.

Masad said “I think ChatGPT is better now,” citing Replit’s internal and publicly posted benchmarks, which he said show two models — Sol and Astra — consistently outperforming on building “vibe-coded” applications from scratch. He said the move away from Anthropic came down to cost, partnership terms, and performance.

Masad’s account was disputed by people familiar with the Replit-Anthropic relationship, however. One person, who asked not to be named, said the partnership was still in place. TechCrunch found evidence on Replit’s website, blog, and social media that the company was using Anthropic’s models as recently as a few weeks ago: a blog post from late September cited Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent’s production environment, and a Replit product employee posted on X around the same time about using Anthropic’s flagship Opus 5.5 model. TechCrunch has asked Replit for clarification and will update the story if it hears back.

Masad said Replit has not yet disclosed annual recurring revenue figures, but pointed to enterprise contracts making up roughly 2% of revenue last year and about 15% this year, with more than 50% of revenue now coming from B2B customers.

Masad also weighed in on AI safety, comparing the requirements for AI agents to those for human workers. He said agents need a “habitat” with good orchestration, error recovery, security, sandboxing, networking, and identity and authorization — and that once those pieces are in place, AI can be deployed safely.

Why it matters

A public, contested break with a major model provider complicates Replit’s story just as rivals court the same developer base — and the conflicting accounts suggest at least one side’s version of the Anthropic relationship doesn’t hold up.