Analysis
Replit CEO and co-founder Amjad Masad said Thursday, onstage at TechCrunch Disrupt, that Replit is no longer working with Anthropic, citing "a number of different reasons, including cost," and adding: "It was a decision based on both cost, partnership, and performance," according to TechCrunch. On performance specifically, Masad said "I think ChatGPT is better now" and pointed to Replit's internal benchmark, where he said Sol and Astra "perform better on building vibe-coded applications from scratch."
A timeline that doesn't fully line up
The announcement is harder to square with Replit's own recent public record than Masad's framing suggests. Replit announced a partnership with OpenAI in August, but TechCrunch found no prior public sign the company had already left Anthropic before Thursday's remarks. Replit's website, a late-September blog post, and a since-posted employee social media update all referenced Anthropic models as recently as a few weeks before the announcement -- the blog post specifically cited Claude Fable 5 and 5.1 running in Replit Agent's production environment. An individual familiar with the arrangement told TechCrunch the Anthropic partnership was, as of that reporting, still in place. TechCrunch said it had reached out to Replit for clarification.
“Pulse has tracked how fast AI coding agents are burning through enterprise budgets industry-wide, Replit included.”
Why the economics matter now
Replit's own numbers show why model costs have become a bigger line item: enterprise revenue grew from roughly 2% of total revenue last year to about 15% this year, and more than half of Replit's revenue is now B2B. That shift toward larger, more cost-sensitive enterprise customers raises the stakes on per-token model pricing in a way that wasn't true when Replit was mostly a consumer and hobbyist coding tool. Masad didn't disclose an ARR figure. TechCrunch notes the Anthropic partnership "was said to have helped the company grow its revenue significantly" (citing Inc.) -- which makes a clean break, if that's what this is, a real bet that in-house models or OpenAI's can hold that growth without Anthropic's help. Pulse has tracked how fast AI coding agents are burning through enterprise budgets industry-wide, Replit included.
What the headline misses
"No longer working with Anthropic" is Masad's characterization, not a documented contract termination, and it's not clear whether it means Replit has fully removed Claude models from its product or is simply routing more traffic elsewhere. Masad also used the moment to argue that AI agents broadly need a "habitat" -- orchestration, error recovery, sandboxing, networking, and identity and authorization -- before they can be deployed safely, a framing that positions Replit's infrastructure as the differentiator regardless of which underlying model is doing the reasoning.
What settles this
Whether Replit updates its own website and documentation to remove Claude references, and whether Anthropic or a source close to the partnership offers a competing account of the relationship's status, will do more to settle this than anything said onstage.
