Replit Says It Dropped Anthropic; Insiders Dispute It

Replit CEO Amjad Masad said onstage at TechCrunch Disrupt on Thursday that his company is no longer working with Anthropic. Insiders familiar with the partnership rejected the claim, and Replit's own blog from late September does not back him up.

What Masad actually said

Masad framed it as a three-part decision, not a falling-out. "It was a decision based on both cost, partnership, and performance," he said. "And for all these reasons, we just felt like they were a better partner for now." On model quality he was blunter: "I think ChatGPT is better now."

He pointed at benchmarks to support it. Replit uses an internal benchmark to help assess metrics like performance, Masad said, and publishes a public one as well: "consistently, Sol and Astra perform better on building vibe-coded applications from scratch." Replit announced a partnership with OpenAI in August, but there had never been any indication that it was no longer with Anthropic.

The pushback

TechCrunch reported that Masad's comments were rejected by insiders familiar with the partnership, with one person who asked not to be named saying the partnership was still in place. The public record points the same way. A Replit blog post in late September cites Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent's production environment. Around the same time, one of Replit's product employees posted on X about using Opus 5.5 and promoting its capabilities. TechCrunch said it had reached back out to Replit for clarification.

Nothing in the reporting settles which account is right. Treat "Replit dropped Anthropic" as one executive's on-stage characterization, disputed by people on the other side of the deal — not as a confirmed contract termination.

The revenue picture behind it

Masad used the same appearance to describe where Replit's money now comes from. Enterprise was "like 2% of revenue last year," he said, while this year "like 15% of our revenue is enterprise contracts, and more than 50% of our revenue is B2B." He declined to share ARR figures. Under the Anthropic partnership, Replit used the lab's Claude models to run its platform, and that partnership was said to have helped the company grow its revenue significantly.

He also argued that agent deployments need infrastructure before they need autonomy: agents, he said, need a "habitat" with "good orchestration, good error recovery, good safety and security and sandboxing, good networking, good identity and authorization."

What this changes if you build on Replit

The practical issue is not which lab wins the argument. It is that on an agent platform you do not pick the model — the platform does, and it can swap underneath you between the day you prototype and the day you ship. Two things are worth doing this week:

• Write down what your app depends on. If a prompt, an eval or a tool-calling pattern was tuned against a specific model generation, note it. A silent backend swap shows up as behaviour drift, not as an error message.
• Read the benchmark as a vendor benchmark. Masad cited an internal benchmark and a public one that Replit itself publishes. That is useful signal about what Replit optimizes for; it is not a neutral comparison, and "building vibe-coded applications from scratch" is a narrow task shape.

What to watch next: whether Replit issues the clarification TechCrunch asked for, and whether Claude models quietly disappear from Replit Agent's production stack the way they were documented in it a few weeks ago.