Replit Swaps Anthropic for OpenAI as Default Model, Muddying a Lucrative Partnership

Replit CEO Amjad Masad delivered a pointed message onstage at TechCrunch Disrupt on Thursday: the AI coding platform is no longer working with Anthropic. The statement, made in front of an industry audience, immediately raised questions about one of the more consequential vendor relationships in the AI coding tools space, and about how much weight to put on a CEO's live commentary versus the fine print of existing contracts.

Masad later softened the framing. Customers can still access Anthropic's Claude models on Replit, he said, but OpenAI is now the default. That distinction matters more than it sounds. Most users never change a default setting. If OpenAI's models now run by default inside Replit Agent, the practical effect is a significant reallocation of inference spend away from Anthropic, regardless of what remains technically "available" on the platform.

The rationale, according to Masad, comes down to three factors: cost, partnership terms, and performance. "I think ChatGPT is better now," he said, pointing to an internal benchmark and a public one that Replit maintains. He specifically cited OpenAI's Sol and Astra models as outperforming rivals on vibe-coded application builds, a use case central to Replit's product. Replit announced its OpenAI partnership in August alongside a free-mode offering, but at the time gave no public indication that it was stepping back from Anthropic.

A Disputed Breakup

The clean narrative, Replit cut Anthropic loose for a better deal, does not survive close scrutiny. People familiar with the Anthropic-Replit relationship told TechCrunch the partnership remains in place, directly contradicting Masad's onstage framing. Replit's own public materials back that up. A blog post from late September references Anthropic's Claude Fable 5 and Claude Fable 5.1 running in Replit Agent's production environment. Around the same period, a Replit product employee posted on social media promoting the capabilities of Anthropic's flagship Opus 5.5 model. None of this reads like a company mid-breakup.

Masad's own walk-back supports the more nuanced version. "Replit remains model agnostic," he told TechCrunch over the weekend. "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." That is a materially different claim than "no longer working with Anthropic." It is a default-model swap framed, onstage, as something more dramatic.

The stakes for Anthropic are real even if the relationship technically continues. Masad said Replit had been, until recently, a top-five customer of Anthropic's, and a previous report noted the Claude partnership helped Replit grow revenue roughly tenfold. Losing default status on a platform of that scale is not a rounding error, even if Claude access persists as an option. Inference economics in coding tools tend to concentrate around whatever model ships by default, since switching requires a deliberate user action that most customers skip.

Why the discrepancy between Masad's stage comments and what insiders describe as an intact partnership? One possibility is simple imprecision, a CEO speaking loosely in a high-pressure interview format and clarifying afterward. Another is that the comments reflect negotiating leverage, a public signal to Anthropic that Replit has options and is prepared to exercise them. Either way, the episode illustrates a structural risk for AI infrastructure providers: customer loyalty is only as durable as the next model that beats a benchmark, and public statements about vendor relationships can shift before the underlying contracts do.

Beyond the model dispute, Masad used his Disrupt appearance to outline Replit's broader business trajectory. Enterprise revenue, he said, has grown from roughly 2% of total revenue last year to about 15% this year, with B2B now accounting for more than half of Replit's business. Backstage, he told TechCrunch he expects enterprise and government customers to generate more than half of Replit's revenue in 2027. Replit has not disclosed ARR figures.

Masad also addressed AI safety, framing it through an operational lens rather than a policy one. He compared AI agents to human workers needing the right environment: "good orchestration, good error recovery, good safety and security and sandboxing, good networking, good identity and authorization." Once that infrastructure is in place, he argued, safe deployment follows. It is a pragmatic, infrastructure-first take on safety that aligns with Replit's identity as a platform company rather than a frontier lab.

The Bottom Line

The headline claim, that Replit has cut ties with Anthropic, does not hold up against the evidence of recent blog posts, employee commentary, and insider accounts. What is more credible, and more interesting from a business standpoint, is the quieter story: Replit has shifted its default model to OpenAI for cost and performance reasons, which will meaningfully reduce Anthropic's share of Replit's inference spend even without a full severance. For Anthropic, the loss of top-five customer status at a platform it reportedly helped scale tenfold is a tangible setback, whatever the technical access terms say. For investors and partners watching the broader AI coding tools market, the real lesson is about switching costs: when the underlying product is modular and model-agnostic by design, loyalty to any single AI lab is conditional on continued benchmark performance, not infrastructure lock-in. Replit's enterprise growth, from 2% to 15% of revenue year over year, suggests the platform itself is gaining durability independent of which model sits behind the curtain. That shift in leverage, from model providers to platform owners, is the more consequential trend here.