Replit makes OpenAI its default agent model; Anthropic ties disputed

Replit CEO Amjad Masad said on stage at TechCrunch Disrupt that the company is no longer working with Anthropic, according to a TechCrunch report originally published on 16 October 2026. Masad said the decision rested on "cost, partnership, and performance". After the talk he clarified that customers can still use Anthropic's models, but that OpenAI's models are now the default. Everything below is TechCrunch's account; I have not seen Replit's own announcement.

What Replit says

• Masad said Replit uses an internal benchmark, and also publishes a public one, and that OpenAI's models "consistently" perform better at building applications from scratch. TechCrunch gives no scores.
• He told TechCrunch that "Replit remains model agnostic" and that OpenAI is "now our preferred partner for powering Replit Agent, based on performance, economics, and the strength of our partnership."
• Masad said Replit had until recently been a top-five Anthropic customer, per TechCrunch.

What is disputed

TechCrunch says people familiar with the relationship rejected Masad's on-stage framing, and one said the partnership was still in place. The outlet also found Replit material from late September that named Anthropic models as running in Replit Agent's production environment. TechCrunch added that Replit announced a partnership with OpenAI in August. Anthropic has not commented in the piece.

Why the default matters

TechCrunch notes that most customers keep the default model rather than choosing one, so the change could cut Replit's spend with Anthropic sharply. For a team buying or building on an agent product, the same logic applies in reverse: the vendor's default decides your model, your bill and your behaviour changes, whatever the model picker offers.

What is not known

• The scores behind the internal or public benchmark, and which models were compared.
• How much cheaper the OpenAI default is for Replit, or whether any saving reaches customers.
• Whether Anthropic's relationship with Replit has ended or only changed.

What a team can take from it

If your coding agent is a third-party product, ask the vendor which model is the default today, how it tells you when that changes, and whether you can pin one. Re-run your own task set after any default change. A vendor benchmark tells you what the vendor measured, not what your repository will do.