Replit makes OpenAI its default over Anthropic, CEO says

Replit's CEO, Amjad Masad, said onstage at TechCrunch Disrupt that the company was no longer working with Anthropic. He pointed to cost, performance, and the partnership itself. Later he softened that: customers can still pick Anthropic's models, but OpenAI is now the default and the preferred partner powering Replit Agent.

The softer version still has teeth. Masad said Replit had been, until recently, one of Anthropic's top five customers. TechCrunch notes most users stay on whatever the default is, so this switch could take a big bite out of what Replit spends with Anthropic, even with Claude still on the menu.

It also says something about the coding-agent market. The big platforms are now willing to move their main model supplier when the numbers change.

Our read is that this is a business decision wearing a benchmark. Masad gave three reasons in the same breath: cost, partnership, and performance. He says Replit's internal and public benchmarks keep favoring the models he named for building apps from scratch. We have no reason to doubt that those benchmarks exist. But a public benchmark run by the company choosing the vendor is not a neutral referee, and this deal began with an August partnership announcement. Economics and alignment almost certainly carried as much weight as raw output.

That's not a knock. Buyers should switch suppliers when price and quality change. If Replit's users get comparable or better apps for less, that's the market working. A model-agnostic platform that actually moves when the numbers move is exactly the pressure that keeps frontier labs competing on price.

The messy part is how it came out. A CEO said on a stage that a partnership was over. Then he walked it back to "remains model agnostic." Meanwhile, a person familiar with the relationship told TechCrunch the partnership is still in place. As recently as late September, Replit was promoting Claude models running in production. Builders who rely on these tools deserve clearer signals than that. When the model under an agent changes, generated code, error patterns, and costs can change with it, and the people shipping on Replit carry that risk.

The bigger thread is concentration of leverage in the defaults. Once a few agent platforms sit between millions of builders and the labs, a single settings change can move a top-five customer's worth of spend overnight. That's good for price competition. It also means the labs will fight hard over distribution deals, and users may not always see why their default model changed.

I'm also watching Replit's own shift upmarket. Masad said enterprise went from about 2% of revenue last year to roughly 15% this year, with more than half of revenue now B2B. He expects enterprise and government to pass half of revenue in 2027. Enterprise buyers ask harder questions about which model touches their code and why. A default chosen partly for partnership reasons will need a cleaner explanation in those rooms.

Replit sells a browser-based platform where users build and deploy apps, increasingly through its AI agent. Before the OpenAI deal, Replit ran on Anthropic's Claude models, a relationship credited with helping its revenue grow. Masad also argued at Disrupt that agents need a proper habitat, meaning orchestration, error recovery, sandboxing, identity, and authorization, before they can be deployed safely.