Replit Drops Anthropic as Its Coding Engine, CEO Says

Replit is no longer working with Anthropic, according to the AI coding platform's co-founder and CEO, Amjad Masad. Speaking onstage at TechCrunch Disrupt on Thursday, Masad said the decision came down to three things: cost, partnership and raw performance.

"I think ChatGPT is better now," Masad said, adding that Replit runs an internal benchmark alongside a public one. "Consistently, Sol and Astra perform better on building vibe-coded applications from scratch. It was a decision based on both cost, partnership, and performance. And for all these reasons, we just felt like they were a better partner for now."

Replit announced a partnership with OpenAI in August, framed at the time as an expansion rather than a replacement.

What Replit's own site shows

The timeline is messier than a clean vendor switch. Insiders familiar with the arrangement disputed Masad's account, with one person telling TechCrunch, on condition of anonymity, that the Anthropic partnership was still in place.

A late-September post on Replit's blog cites Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent's production environment, and around the same time a Replit product employee posted on X about using Anthropic's flagship Opus 5.5.

Under the earlier arrangement, Replit used Claude models to run its browser-based development and deployment platform, a partnership credited with helping the company grow revenue significantly. Replit did not dispute that history, only its present tense.

Why the split matters for builders

For most of the last two years, the assumption inside AI-native products was that the model layer was sticky: choose a lab, build around its quirks, grow together. The Replit episode suggests the opposite. A coding platform with millions of users can apparently swap its reasoning engine the way a SaaS company swaps a payments provider, on the strength of an internal benchmark and a per-token invoice.

That has two consequences. The first is leverage: every serious coding tool now negotiates against at least three frontier vendors, and the threat of switching is credible because switching has been done. The second is fragility. When a platform's docs, its employees' public posts and its CEO's stage comments disagree about which models are in production, users cannot tell what they are paying for.

Masad said enterprise moved from roughly 2 percent of Replit's revenue last year to about 15 percent this year, with more than half of revenue now coming from business customers. Enterprise buyers ask for model provenance, data handling terms and fallback plans. A vendor relationship that changes on a conference stage, before the docs catch up, is what procurement teams now put in writing.

Masad's own checklist for what agents need is a good summary: "good orchestration, good error recovery, good safety and security and sandboxing, good networking, good identity and authorization." None of that comes from a model vendor alone. Which is the argument for keeping the model layer replaceable.