Replit drops Anthropic for OpenAI, CEO says cost and performance drove it

Replit is no longer working with Anthropic, CEO and co-founder Amjad Masad said Thursday while speaking onstage at TechCrunch Disrupt. The break ends a model partnership that reportedly helped the AI coding platform grow its revenue significantly, and it lands just weeks after Replit's August partnership announcement with OpenAI.

Until now, Replit never signaled that its Anthropic relationship had ended. The company sells a cloud-based development platform where users build and deploy apps directly from a browser, and under the Anthropic deal it ran on Claude models.

Why did Replit switch?

Masad pointed to three factors: cost, partnership terms, and raw model performance. He backed the performance claim with Replit's own testing.

"I think ChatGPT is better now," Masad said. "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. And for all these reasons, we just felt like they were a better partner for now."

Replit uses an internal benchmark alongside its public one to assess metrics like performance, Masad explained. The phrase "for now" is worth noting — it frames the choice as provisional, not a permanent lock-in.

The takeaway: Replit says it made a business decision on price and benchmarks, not a loyalty play.

Is the Anthropic partnership actually over?

Not everyone agrees with Masad's account. One person familiar with the Anthropic-Replit partnership, who asked not to be named, said the partnership was still in place.

There's also technical evidence that Replit kept using Anthropic products as recently as September. A Replit blog post from that month cites Anthropic's Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent's production environment.

That leaves a gap between the CEO's onstage statement and the company's own recent documentation. TechCrunch reached back out to Replit for clarification and said it will update its story if it hears back.

The takeaway: the split may be partial, recent, or messier than a clean breakup sounds.

What's driving Replit's revenue now?

Masad used the Disrupt stage to sketch how the business mix has shifted. Enterprise is becoming a much bigger slice.

"We're not sharing ARR numbers just yet, but in terms of growth, like, I would say, enterprise was like 2% of revenue last year," Masad said. "This year, I would say like 15% of our revenue is enterprise contracts, and more than 50% of our revenue is B2B."

That's a fast shift for one year:

• Enterprise contracts: 2% of revenue last year → 15% this year
• B2B overall: more than 50% of revenue
• ARR: not disclosed

A platform that started as a developer favorite is now leaning on corporate buyers. Model-partner decisions like this one carry more weight when enterprise customers are watching reliability, cost, and vendor relationships.

The takeaway: Replit's customer base is changing, and its AI suppliers have to keep up with that.

What did Masad say about AI safety?

Masad also argued that deploying AI agents safely requires building the right environment around them — what he called a "habitat."

"In the same way that us humans need a certain environment to work — you need a nice office, maybe nice good food in the office, air conditioning, computer — agents need a habitat for them to exist and do good work," he said. "So that habitat has good orchestration, good error recovery, good safety and security and sandboxing, good networking, good identity and authorization. And once you figure all of that out, then you can deploy AI safely."

His checklist for that habitat:

• Good orchestration
• Good error recovery
• Good safety, security, and sandboxing
• Good networking
• Good identity and authorization

The takeaway: in Masad's framing, safe AI deployment is an infrastructure problem first.

Why does this split matter?

AI coding tools are one of the most competitive corners of the market right now, and they're also some of the biggest customers for foundation-model labs. When a platform like Replit moves its inference business between Anthropic and OpenAI, it shifts real spending and sends a signal about which models win on coding tasks.

The stakes cut both ways. For Anthropic, losing a customer that credits the partnership with significant revenue growth stings. For OpenAI, landing Replit in August and converting it into its primary partner is a marquee win in the developer-tools segment.

And the confusion around the announcement matters too. A September blog post referencing Claude Fable 5 and Claude Fable 5.1 in production, plus a source saying the partnership continues, suggests the transition may still be in progress. Companies building on frontier models increasingly run multi-provider setups, switching between models based on cost and benchmark results — exactly the behavior Masad described with his "for now" phrasing.

The takeaway: expect more coding platforms to treat model providers as interchangeable suppliers, renegotiated on benchmarks and price.

What happens next?

Replit hasn't responded yet to the request for clarification, so the exact status of the Anthropic relationship — fully terminated or winding down — remains unclear. What's certain is the direction: Masad says OpenAI's models now win on Replit's internal and public benchmarks for building "vibe-coded" apps from scratch, and the company is paying accordingly. Whether Anthropic's next model cycle flips that calculus back is the question both labs are now competing to answer.