Replit has made OpenAI the default engine behind its coding agent, pulling back from Anthropic in a switch its chief executive described in unusually blunt terms. Speaking on stage at TechCrunch Disrupt on Thursday, co-founder Amjad Masad said the company was no longer working with Anthropic and had moved Replit Agent over to OpenAI's models. He cited a mix of cost, performance and the shape of the partnership.
The remark landed harder than a routine vendor change, and for a reason. Anthropic built much of its commercial reputation on code. Claude became the model many developers reached for first, and a steady stream of coding tools wired themselves to it. A platform like Replit, which puts an AI agent at the centre of how people build software, choosing a rival as its default is the kind of signal the market reads closely.
The practical change is smaller than the headline. Claude has not been removed. Replit users can still select Anthropic's models by hand, so the switch is about what runs when nobody changes the setting. That default matters more than it sounds, because most users never touch it. If the bulk of Replit's traffic now flows to OpenAI, Anthropic quietly loses a meaningful slice of usage whether or not the two companies are formally done.
There is some disagreement about how clean the break really is. A person familiar with the relationship told TechCrunch that Masad had overstated things, and Replit's own pages and staff were still referencing Claude after the comments. The most careful reading is the narrow one: Replit has changed its default provider, which is not quite the same as tearing up every tie to Anthropic.
Cost is the part worth sitting with. Anthropic has been spending enormous sums to secure compute, including a roughly $60 billion debt arrangement to lock in chips. Those bills eventually reach customers through token prices, and a company running an agent that burns tokens by the million feels every cent. Masad's reasoning reads less like a grudge than a spreadsheet.
It also fits a pattern that has not been comfortable for Anthropic. The lab recently lost a high-profile government user when the Pentagon stepped back from Claude, and now a well-known developer platform has reset its default. Neither is fatal on its own. Anthropic's enterprise business is large and growing, and it keeps shipping, including the cheaper Haiku 5.5 aimed squarely at cost-sensitive workloads. But defaults are sticky, and every one that flips to a competitor is a reminder that loyalty in this market tends to last exactly as long as the price and the benchmarks hold.
Commentarii · 0