Replit Ends Anthropic Deal, CEO Cites Cost as Coding Platform Deepens OpenAI Ties

Replit has stopped using Anthropic's models, with CEO Amjad Masad pointing to cost among several reasons, months after the company announced a partnership with OpenAI.
Replit is no longer working with Anthropic, CEO Amjad Masad said, confirming that the AI coding platform has dropped the Claude maker's models from its stack. Masad attributed the split to multiple factors, with cost leading the list. The disclosure is notable because Replit never signaled that the relationship had ended, even after it announced a partnership with OpenAI in August.
The two companies had been closely linked throughout the coding-assistant boom, when Anthropic's Claude models became a default choice for tools that write and debug software. Replit's move shows how quickly those allegiances can shift once a rival offers better economics or a tighter product fit. For Anthropic, it removes a visible deployment inside one of the better-known browser-based development environments.
Cost is the stated driver, and the math is easy to see. Agentic coding products fire off long chains of model calls, so inference bills scale with every user session rather than with a fixed subscription. A platform serving millions of developers has strong incentive to push that traffic toward whichever provider offers the best price per token, or to negotiate volume terms that a smaller vendor cannot match.
The timing lines up with Replit's August tie-up with OpenAI, which gave the company a new model partner and a public-facing alliance. Until now, that announcement read as an addition rather than a replacement. Masad's comments reframe it: the OpenAI deal appears to have displaced Anthropic's models rather than sitting alongside them.
For developers who build on Replit, the practical effect depends on which models sit behind the features they use. Agent and autocomplete quality can shift when the underlying model changes, even if the interface looks identical. Teams with Replit embedded in their workflow may want to re-test output on their own codebases rather than assume parity with earlier sessions.
Multi-model strategies are becoming the norm across coding tools, and Replit is not the first to swap suppliers. The open question is whether Anthropic loses more ground among developer platforms as OpenAI and Google compete on price. Watch whether Replit names the OpenAI models now powering its agent, and whether other coding startups follow with similar disclosures.