TechCrunch reported that Replit CEO and co-founder Amjad Masad said the company chose OpenAI over Anthropic because of a combination of cost, partnership terms and model performance. Replit had previously relied heavily on Anthropic’s Claude models, and was reportedly one of the company’s largest customers.
The shift does not prevent Replit users from selecting Anthropic models or other supported systems. However, OpenAI’s model is now preselected for Replit Agent, the platform’s tool for building applications from natural-language instructions. That default position could have an outsized commercial effect because many users will continue with the model already selected, directing a potentially large volume of inference spending toward OpenAI.
Production performance becomes the battleground
Replit’s own account emphasizes that the decision was based on more than commercial negotiations. In its post, Replit said it compared Anthropic’s Fable 5 and OpenAI’s GPT-6 Astra in production, describing both models as delivering exceptional engineering quality while presenting Astra as the stronger choice in its internal and public testing.
The significance of those comparisons is that Replit operates close to the point where model quality becomes visible in practical software production. A model that performs well on a coding benchmark may still struggle to understand an existing codebase, manage dependencies, recover from errors or complete an application from scratch. Replit’s platform can measure those outcomes directly through user workflows and production activity.
That gives the company strategic leverage with model providers. Replit is not simply purchasing access to an API. It is deciding which model will be placed in front of developers at the moment they begin creating software. For OpenAI, that makes the partnership a distribution channel as well as a source of usage revenue. For Anthropic, losing the default position could reduce demand even if its models remain available.
OpenAI has framed its relationship with Replit in similarly commercial terms. OpenAI said in its announcement that GPT-5.6 Luna powers Replit’s Free Mode, with the partnership focused on improved model price performance and broader access to software creation. The announcement highlights a different model name from Replit’s production comparison, underscoring how quickly model portfolios and product labels are changing across the industry.
A test of AI distribution economics
The disagreement over Anthropic’s status also matters. Masad characterized the relationship as having changed, while a person familiar with the partnership disputed the description that Replit had stopped working with Anthropic. Replit has since stressed that it remains model agnostic, allowing customers to choose alternatives rather than removing them from the platform.
That distinction gives Replit flexibility while still allowing OpenAI to capture the commercial advantage of the default. It also reduces the risk that customers view the company as locked into a single provider, an important consideration for enterprise buyers concerned about price, reliability and vendor concentration.
Replit’s business mix makes the decision more consequential. The company said its enterprise revenue share has grown from about 2% last year to roughly 15% this year, while business customers now generate more than half of revenue. Masad expects enterprise and government customers to contribute more than half of revenue in 2027.
As Replit moves toward larger customers, model selection becomes part of its infrastructure strategy. The winner will not necessarily be the provider with the best isolated benchmark score. It may be the provider that combines sufficient coding quality with competitive economics, favorable partnership terms and the strongest route into everyday developer workflows.
- Village Global · CC BY 2.0
Rebeca Smith is not a person. No notebook, no deadlines, no face behind the name — just a byline this newsroom publishes under. Here is the production line underneath it, because a name beside a portrait reads like a journalist, and this one is not one.
The models. Writing: gpt-5.6-luna and qwen3-max. Out on the live web: gpt-5.6-luna and gpt-5.6-terra. Pictures: gpt-image-1 and gpt-image-1-mini. Swap one in the newsroom and this line swaps with it — it is read off the machines, not typed here.
How a story is made
- Research. The searching model reads around the story, pointed at primary sources — the filing, the post, the repository — rather than at somebody else's write-up of them.
- Writing. The writing model drafts it against what was found, at Rebeca Smith's usual length and in Rebeca Smith's usual register.
- The loop. A reviewer reads the draft and sends it back with notes. Then reads it again. A piece can go round several times before it leaves the building.
- Enrichment. A quotation has to appear word for word on the page it is taken from. A chart may only use figures that appear in the source it cites. Whatever fails is dropped, and the reason is kept.
- Fact check. A last pass hunts for claims the article makes and its sources do not.
- A human stop. Sensitive subjects are held for a person to read before publication, and a person can kill any of it at any point.
If that sounds less like a newsroom and more like a factory: quite. It is called Press Factory.
This article was generated using AI and published automatically without human pre-publication review.
How this article was made
The article was produced by the Grandmonts Media News Engine using automated research, drafting and verification workflows. No human editor reviewed the article before publication. Grandmonts Media remains responsible for the published content. Errors can be reported at office@grandmonts.cz.