Replit CEO Says Company No Longer Works With Anthropic, Then Says Claude Is Still Available as OpenAI Becomes Default

Replit CEO Amjad Masad said onstage at TechCrunch Disrupt on Thursday, Oct. 15, that his company is no longer working with Anthropic. Within days, the picture got narrower. Anthropic's Claude models are still on the platform. OpenAI is now the default.

The gap between those two statements is the story.

What Masad said

Asked by the interviewer whether Replit's August partnership with OpenAI meant it was done with Anthropic, Masad answered, "Yeah. I mean, yeah."

He gave three reasons: cost, partnership and performance. "I think ChatGPT is better now," Masad said. He pointed to an internal benchmark and a public one Replit publishes. "Consistently, Sol and Astra perform better on building vibe-coded applications from scratch," he said. OpenAI is "a better partner for now."

Masad also said Replit had been, until recently, a top-five Anthropic customer. The interviewer had asked whether losing Replit contributed to what the interviewer described as a reported Anthropic revenue slowdown.

The walk-back

After Disrupt, Masad clarified that customers can still pick Anthropic models on the platform. "Replit remains model agnostic," he told TechCrunch. "Our users can choose Anthropic's Claude or other supported models. OpenAI is now our preferred partner for powering Replit Agent, based on performance, economics, and the strength of our partnership."

That is a default switch, not a severed relationship. It still matters financially. Most customers stick with whatever the default is, so the change could cut Replit's spending with Anthropic sharply. Replit has not published what it spends or by how much.

What Replit's own materials show

Replit's model-selector documentation lists seven Claude models as Max Mode options, including Claude Fable 5.1, Claude Opus 5.5 and Claude Sonnet 5. They sit alongside GPT-6 Astra, GPT-6 Luna Fast and Kimi K3. GPT-6 Sol is offered in Power Mode.

A Sept. 29 engineering post on Replit's blog says Claude Fable 5 ran in Replit Agent's production environment in August. It says Fable 5.1 ran alongside GPT-6 Astra in September. The post reports that Replit Agent, with Astra as its core loop in Max mode, scored 72% at $2.11 per task on the DeepSWE v1.1 benchmark. It scored 49% at $2.53 per task on Terminal-Bench 4.0. These are Replit's own figures.

Around the same time, a Replit product employee posted on X about using Anthropic's flagship Opus 5.5 and praising its capabilities.

The dispute over "no longer working with"

One person familiar with the Replit-Anthropic relationship, who asked not to be named, told TechCrunch the partnership is still in place. Masad's onstage description was rejected by that person.

Masad's position, in his own later words, is that Replit is model agnostic and has made OpenAI its preferred partner for the Agent. Replit's documentation is consistent with both accounts. Claude models are listed and selectable, and OpenAI's are now the default. The documents don't show whether a commercial agreement with Anthropic remains in force.

The September blog post also complicates the performance claim in one respect. It credits the code quality of both Astra and Fable 5.1 for letting Replit's engineers use their code-review subagent less, with no drop in eval results. Masad's onstage case was that OpenAI's models are simply better. The engineers' post credits both model families.

The business behind the switch

Masad used the Disrupt session to pitch Replit as a full-stack environment, not just a code generator. He said OpenAI and Anthropic are moving into building, deploying and hosting apps, but that those offerings remain surface level. He noted that OpenAI relies on an external service, Cloudflare, for hosting.

Replit's agents, he said, can spin up databases and key-value stores inside a production environment. The company can deploy its whole stack into government and enterprise environments on-premises, he added.

He also gave a rough revenue picture. Enterprise was about 2% of Replit's revenue last year, and this year it is about 15%, he said. More than 50% of revenue is B2B. Replit is not sharing annual recurring revenue figures yet.

What is still open

Replit has not disclosed the terms of its OpenAI deal or how much its Anthropic spending changes. The company also hasn't said whether it still has a commercial agreement with Anthropic. That is the exact point Masad's onstage answer and the anonymous account contradict.

Users will see the effect first. Replit Agent sessions now route to OpenAI models by default, while Claude stays one click away in the model selector.