AI Coding Platform Dumps Anthropic for OpenAI, Says ChatGPT Models Cheaper and Better

Replit has made OpenAI its preferred AI model provider, with CEO Amjad Masad saying its models offer better performance at a lower cost for the platform’s needs. But his initial claim that Replit was no longer working with Anthropic has since been qualified: customers can still access Claude models, and a source familiar with the partnership disputes that the relationship has ended.

Speaking at TechCrunch Disrupt on Thursday, Masad said Replit was “no longer working with Anthropic.” Asked whether the decision followed Replit’s OpenAI partnership announced in August, he replied, “I mean, yeah.”

Masad cited cost, model performance and the strength of the OpenAI partnership as reasons for the shift.

“I think ChatGPT is better now,” Masad said. He added that OpenAI’s Sol and Astra models consistently performed better on Replit’s internal and public benchmarks for building applications from scratch, including AI-assisted development often called “vibe coding.”

Those results reflect Replit’s own evaluation methods and workloads. They do not establish that OpenAI models outperform Claude across all coding tasks, and Replit has not disclosed comparable figures showing exactly how much cheaper its preferred models are.

OpenAI Is the Default, but Claude Remains Available

After the event, Masad clarified that Replit customers can still choose Anthropic models, while OpenAI is now the default provider for Replit Agent.

“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.”

The clarification changes the scope of the announcement. Replit has shifted its preferred provider rather than removed Claude from the platform. Customers may continue using a provider without actively selecting it, so changing the default could redirect usage and spending toward OpenAI even if Anthropic’s models remain available.

The Extent of the Anthropic Split Is Disputed

A source familiar with the partnership told TechCrunch that the relationship remained in place, disputing Masad’s initial description of the split.

Replit’s own engineering updates provide evidence that Anthropic models were still being used recently. A September post described Claude models running in Replit Agent’s production environment, while a Replit employee also discussed using Anthropic’s Opus model.

In a separate assessment of its production models, Replit’s engineering team wrote: “The code quality of Astra and Fable 5.1 also let us use our code-review subagent less, with no drop in our eval scores.”

These details do not establish the current commercial terms of the partnership or how extensively Replit continues to rely on Anthropic’s technology behind the scenes. Neither does continued model availability prove that the two companies maintain the same business arrangement as before.

For now, the clearest confirmed change is that OpenAI has become Replit’s preferred provider for powering Replit Agent. The claim that the companies have completely parted ways remains disputed.

Why the Shift Matters to Both Companies

Replit announced its OpenAI partnership in August after previously relying on Anthropic’s Claude models to power parts of its platform. Masad said Replit had been one of Anthropic’s top five customers until recently, suggesting the change could be significant even without a complete termination of the relationship.

For OpenAI, becoming Replit’s default provider offers a way to reach users through an established AI coding platform. For Anthropic, losing default status at a customer that was reportedly among its largest could reduce usage and revenue, although the financial impact has not been disclosed.

The cost argument also remains unquantified. Masad says OpenAI’s models perform better and cost less for Replit’s workloads, but without comparable pricing figures and independently reproducible benchmark results, outsiders cannot assess the size of the advantage.

Masad also told TechCrunch that he expects enterprise and government customers to account for more than half of Replit’s revenue in 2027. As the company pursues larger business customers, model performance, operating costs and the strength of its provider partnerships could become increasingly important.

The next indicators to watch are how usage shifts between providers, whether Replit discloses more detail about its model costs and whether either company clarifies the status of their commercial relationship.