Replit has officially pulled the plug on its partnership with Anthropic, a move laid bare by CEO and co-founder Amjad Masad at TechCrunch Disrupt. The cloud development platform is ripping out and replacing its underlying model stack right in the middle of its operational scaling, trading suppliers based purely on the harsh economics of inference. As Amjad Masad stated, the pivot comes down to a brutal calculation of system costs and real-world performance.
Masad said the decision was made for a number of different reasons, including cost.
Such infrastructure gymnastics expose the cutthroat scramble for operating margins across the wrapper platform ecosystem. As recently as August, the company loudly touted its integration with OpenAI while hedging its bets with competing models. Yet the relentless gravity of inference costs during application generation forces platforms to jettison exclusive vendor allegiances in favor of ruthless pricing discipline and internal benchmarks.
Internal Metrics Versus Model Provider Ambitions
Replit justifies ditching battle-tested architectures by citing its own proprietary testing regimes. As Amjad Masad pointed out, the company's internal benchmark exposes the clear superiority of its newly adopted routing for greenfield software assembly.
“I think ChatGPT is better now,” Masad said, adding that the company uses an internal benchmark to help assess metrics like performance.
The platform is now championing its own public metrics under the codenames Sol and Astra, which management claims consistently smoke the competition on autonomous coding tasks. This sets a fascinating market precedent: wrapper platforms are starting to dictate terms to foundational model builders, leaning entirely on their own unit economics. At the same time, industry observers note a glaring gap between executive posturing and the company's recent technical deployments, which relied heavily on Claude Fable 5, Claude Fable 5.1, and the flagship Opus 5.5 inside Replit Agent environments.
Financial Restructuring and Revenue Mix Shift
Replit's financial disclosures reveal a violent pivot toward the enterprise tier. According to Amjad Masad, enterprise contracts accounted for roughly 2% of revenue last year, but have ballooned to 15% this year, with total B2B sales surging past 50%.
Scaling contracts of this magnitude demands predictable code generation costs and an iron grip on infrastructure burn. When inference bills threaten to devour the margins of B2B software delivery, architectural decisions stop being about technical purity and start answering directly to the balance sheet. For any enterprise leadership team watching this play out, the directive is clear: audit every single vendor SLA for hidden volume penalties before the current fiscal quarter closes.