Replit says it has stopped working with Anthropic, but partnership status is disputed

Replit chief executive Amjad Masad says the AI coding company has stopped working with Anthropic, citing cost, partnership terms and the performance of OpenAI’s models. But TechCrunch also reports that people familiar with the arrangement dispute that it has ended, with Claude models apparently still in use at Replit. The split between those accounts matters more than a neat breakup headline would suggest.

Watch Desk analysis

What happened

Masad made the announcement at TechCrunch Disrupt 2026, saying OpenAI’s models now perform better for Replit’s application-building work. TechCrunch reports that people familiar with the partnership say the agreement remains in place, and points to engineering records and product integrations suggesting Claude is still being used.

So the public picture is unsettled: Replit’s CEO says the company is no longer working with Anthropic, while sources familiar with the partnership dispute that account. The available details do not resolve whether this means a formal end, a change in the work being done, or something less tidy.

Why it matters

This is a live question about which models power a coding platform, not just a bit of vendor musical chairs. Replit’s explanation puts cost, commercial terms and performance in the frame; the reported continuing Claude use suggests the practical transition may be less clean than the announcement implies. For developers, the distinction could matter if model availability or behaviour changes in the product they use.

The disagreement is also a useful reminder that a company’s public account of a partnership and its technical integrations do not always move in lockstep. That is worth tracking, without treating either a claimed departure or continued use as settled fact.

Our read

The headline is not yet “Replit dumps Anthropic” so much as “Replit says one thing, people familiar with the deal say another”. That is less punchy, but rather more faithful to the evidence. The useful test will be whether Replit confirms a change in the models powering its tools, and whether Claude integrations actually disappear.

What to watch

• Whether Replit or Anthropic clarifies the partnership’s current status.
• Whether Replit changes which models are available for building applications.
• Whether Claude integrations remain in use or are removed.

Discussion spark: If a company says a partnership has ended but its product still appears to use the partner’s models, should the public account or the technical evidence carry more weight?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.