On 17 October 2026, Replit CEO Amjad Masad told TechCrunch Disrupt that the company’s default coding model had moved to OpenAI. On stage he also said Replit was no longer working with Anthropic. Afterwards he clarified that customers can still choose Anthropic models, and that the change is the default. Most people never change a default. If an agent writes your product, that switch is your release process.
What was said, and what stayed fuzzy
Masad cited cost, performance, and the partnership. He said ChatGPT was ahead on Replit’s view of the benchmarks, including tasks that build an application from a prompt, and that Replit had been one of Anthropic’s larger customers. He also said enterprise revenue had grown from about 2 percent of revenue to about 15 percent. People familiar with the companies, in follow-up reporting, disputed a clean break and said the Anthropic relationship was still in place. The durable fact for a buyer is the default, not the stage line. TechCrunch: https://techcrunch.com/2026/10/17/replit-is-no-longer-working-with-anthropic-ceo-says/
Questions to ask any coding agent
- Which model is the default this month, and who decides the next one?
- Can we pin a model for a regulated client, or does “better benchmark” move us without a test on our repo?
- Where does the code live when the vendor changes partners? Your git remote should not be a feature of their plan.
- What do you retest after a model swap? A greener public benchmark can still break your billing flow.
A coding agent is a fast draft. The thing you sell is the repository, the tests, and the person who can explain a change six months later. That is the ordinary custom software development process, with or without a default model in someone else’s IDE.
FAQ
Should we leave Replit because the default changed?
Leave if you cannot pin the model, export the code, or rerun your tests after a swap. Stay if those three are true and the new default passes your suite. The logo on the model is the smaller issue.
Is OpenAI now the safer choice?
It is Replit’s current default, on their evaluation. Your evaluation is your codebase. Run the same tasks on the models you are allowed to select before you standardise.
Keep a human on the merge
Contact Altron Technologies if agents are opening changes faster than anyone is reviewing them. We can put a release path around the draft.