Replit’s CEO says it has dropped Claude for OpenAI because ‘ChatGPT is better now’

Replit, the browser-based app builder that grew up on Anthropic’s Claude models, has stopped working with Anthropic and moved to OpenAI, its chief executive Amjad Masad told the audience at TechCrunch Disrupt on Thursday. The claim is already disputed: a person familiar with the partnership told TechCrunch it is still in place, and neither Replit nor Anthropic has put out a statement.

Masad gave three reasons for the switch: cost, the partnership itself and how the models perform. Replit announced a partnership with OpenAI in August, when it launched a free mode for Replit Agent, but until now there had been no sign that Anthropic was on the way out.

‘I think ChatGPT is better now’

Masad said Replit judges models on an internal benchmark as well as a public one, and that OpenAI’s current models come out ahead on the work Replit’s users do most:

We publish a public benchmark as well, and consistently, Sol and Astra perform better on building vibe-coded applications from scratch. It was a decision based on both cost, you know, partnership, and performance.

Amjad Masad, Replit CEO, at TechCrunch Disrupt

Sol and Astra are OpenAI’s GPT-6 models. GPT-6 Sol arrived in September at half the price of OpenAI’s previous flagship, launched 90 minutes after Anthropic released Claude Opus 5.5, and price is the first reason Masad gave. He said OpenAI was “a better partner for now”, which leaves the door open to a return.

Replit’s own blog still shows Claude at work

Masad’s account doesn’t match what Replit published three weeks ago. A September 29 engineering post on how Replit Agent hands work to cheaper sub-agents compares three models in Replit Agent production, one week each: Anthropic’s Claude Fable 5 in August, and Claude Fable 5.1 and OpenAI’s GPT-6 Astra in September.

The figures in that post favour OpenAI on the behaviour Replit was designing for. GPT-6 Astra handed work to a general worker on 20% of turns, against 2.3% for Fable 5.1 and 0.9% for Fable 5, and Replit wrote that Astra “decides for itself where its tokens are worth spending.” The post doesn’t say Replit has stopped using Claude.

That gap is why the claim is contested. One person familiar with the partnership told TechCrunch it is still in place. TechCrunch has asked Replit to clarify; Anthropic hasn’t commented.

Enterprise is now 15% of Replit’s revenue

Masad also gave a rare look at how Replit’s business is changing. He wouldn’t share annual recurring revenue figures, but said enterprise contracts were about 2% of revenue last year and are about 15% this year, and that more than half of Replit’s revenue now comes from business customers.

Replit made its name with founders and hobbyists building apps by describing them in plain English, the same vibe coding that Anthropic’s own Claude Code is now known for. That puts Replit in an awkward spot: its model suppliers increasingly sell coding tools that compete with it.

Why it matters

Coding tools are a big part of Anthropic’s business, and Replit was one of the best-known companies built on Claude, so a public defection lands just as Anthropic tells IPO investors its growth is cooling. Whether Replit has dropped Claude entirely or is shifting most of its traffic to OpenAI, it shows how quickly an app builder can swap models when one gets cheaper, the point made in our look at the AI labs’ missing moat.

Sources: TechCrunch; Replit, Free the models: Harness design at the frontier; Replit, Replit Introduces Free Mode.