Replit Ends Anthropic Partnership, Picks OpenAI
Replit is done with Anthropic. Or is it? The company’s CEO said so onstage this week, but people close to the partnership tell a different story.
Speaking at TechCrunch Disrupt on October 16, 2026, Replit CEO and co-founder Amjad Masad was asked whether the OpenAI partnership Replit announced in August meant the Anthropic relationship was over. “Yeah. I mean, yeah,” he answered.
He did not hedge on the reasons either. Cost. Partnership terms. Performance. “I think GPT is better now,” Masad said, pointing to Replit’s internal VBench benchmark and a public benchmark the company publishes. He claimed OpenAI’s Sol and Astra models consistently outperform rivals at building vibe-coded applications from scratch.
Why this stings for Anthropic
Replit is not a small customer. Masad himself said the coding platform was one of Anthropic’s top five customers. Under the partnership, Replit ran its browser-based development platform on Claude models, and Masad credited that relationship with driving significant revenue growth.
Losing a top-five customer to your biggest rival is bad. Losing them while they praise your rival’s models on a conference stage is worse.
Masad also shared that enterprise contracts now make up about 15% of Replit’s revenue, up from just 2% last year. More than half of total revenue comes from business customers. For that enterprise push, Replit is betting on OpenAI as the better partner.
The dispute
Here is where it gets messy. After Masad’s comments, insiders familiar with the partnership told TechCrunch the cooperation is still ongoing. Replit’s own September production notes listed Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent’s production environment. A Replit employee discussed using Anthropic’s Opus 5.5 around the same time.
So which is it? The honest answer is that “partnership” can mean many things. Replit may have ended its commercial model-supply deal while traces of Claude linger in testing environments. Or the breakup may be aspirational rather than complete. Neither company has clarified the scope.
What is clear is the direction of travel. Replit announced its OpenAI partnership in August. Its CEO now publicly prefers GPT models. Whatever the paperwork says, the strategic bet has shifted.
What it means for developers
If you build with AI development tools, this is a reminder that the model underneath your favorite platform can change. Benchmarks, pricing, and partnerships all move fast. The tool you use today might run on a different model next quarter.
For Anthropic, the bigger worry is the pattern. When a top customer leaves citing cost and performance in public, competitors notice. And enterprise buyers, the exact segment every AI lab is chasing, pay close attention to who the serious builders choose.
FAQs
Did Replit really end its Anthropic partnership?
CEO Amjad Masad said so onstage at TechCrunch Disrupt on October 16, 2026. However, insiders told TechCrunch the partnership is still in place, and Replit’s September notes still referenced Claude models. The full scope of the split is unclear.
Why did Replit switch to OpenAI?
Masad cited cost, partnership terms, and performance. He said OpenAI’s Sol and Astra models consistently performed better on Replit’s benchmarks for building applications from scratch.
How big was Replit as an Anthropic customer?
Masad said Replit was one of Anthropic’s top five customers, making the loss significant.
