Replit's CEO Said He Dropped Anthropic. He Hadn't.

Replit CEO Amjad Masad told a room at TechCrunch Disrupt this week that his company was "no longer working with" Anthropic, and that OpenAI's models now power its Agent product by default. Within a day, he was walking the claim back on X, and people close to the Anthropic partnership told TechCrunch it never actually ended. For anyone building on Replit, or just watching the OpenAI-versus-Anthropic fight for coding-agent market share, the confusion matters more than the walk-back suggests.

What changed

Masad made the comment on stage at TechCrunch Disrupt on October 15, citing cost, the strength of the OpenAI partnership, and Replit's own internal benchmark, where he said OpenAI's models "consistently" outperformed Claude on building vibe-coded apps from scratch. Asked directly whether that made sense given Claude Code's following among developers, he said: "I think GPT is better now."

A day later, Masad posted a clarification on X: Replit "remains model agnostic," Claude is still available to users, and what he meant on stage was that OpenAI is now the "preferred partner" powering Agent by default. People familiar with the Anthropic deal went further, telling TechCrunch the partnership is still active and that "no longer working with" overstated things. Replit's own blog, as recently as late September, named two Claude models running inside Agent's production environment. Masad has previously said Replit was a top-five Anthropic customer.

techcrunch.com/2026/10/17/replit-is-no-longer-working-with-anthropic-ceo-says/

The timing is odd on its own terms: Replit announced an OpenAI partnership back in August with no suggestion Claude was on its way out, and the same week as this story, Replit separately launched Bring Your Own Cloud for enterprise customers, letting teams run Agent's output on their own AWS, Azure, or Google Cloud infrastructure.

Why it matters

Strip away the back-and-forth and one fact stands: Replit changed its default model provider, and almost nobody switches a default. If OpenAI is now what a new Replit Agent session reaches for out of the box, that's a real shift in Anthropic's revenue from one of its highest-profile agent customers, regardless of whether Claude stays selectable in a dropdown. The whole argument on stage was framed as a question of which model is better. What actually changed was a billing default, and that's a different question with a different answer.

It's also a preview of a fight that's going to keep happening. Every AI coding platform that wraps multiple model providers, Replit, Cursor, Windsurf, the agent nodes inside n8n workflows, eventually has to pick a default, and that choice is made on cost and internal benchmarks that outsiders never see. Masad's framing, that developer sentiment toward Claude Code doesn't translate into which model a platform actually ships by default, is the uncomfortable part a press release wouldn't say out loud.

What people are saying

Karl Mehta, who was in the room at TechCrunch Disrupt, posted the exchange that set off the story: Masad said Replit chose OpenAI "on cost, partnership and performance," and that asked about Claude Code's following, he said "I think GPT is better now."

Abhishek Madhavan called the on-stage remark "a big statement" given how central Agent's output is to Replit's product, and noted the conflict with TechCrunch's own sourcing: "A source told TechCrunch the partnership is still in place."

And Masad's own follow-up post is really the third voice in this story, since it reads as a correction of his own words: "To clarify: Replit remains model agnostic. Our users can choose Anthropic models as well as many other supported models. What I meant at TC Disrupt is that OpenAI is now our preferred partner powering Agent based on performance, cost, and strength of our partnership."

Nobody from Anthropic has responded publicly as of this writing.

What it looks like in practice

Before this week, opening a new Replit Agent session meant Claude handled the work unless you changed it. Now the same new session reaches for an OpenAI model first, and Claude sits one dropdown click away. Nothing about what you can build changed. What changed is which company gets paid for the average session, since most users never touch that setting.

It's the same mechanic n8n builders deal with inside the AI Agent node: pick a model provider once when you wire up the workflow, and that choice runs quietly in the background for every execution after, often for months before anyone checks whether a newer or cheaper model would do the job better. Replit's situation is just that decision made at platform scale, in public, by a CEO who then had to explain what he meant.

For context on how close the OpenAI-versus-Anthropic coding race actually is right now, see our coverage of OpenAI cutting Cursor off from future models and the head-to-head reliability numbers between Opus 5.5 and GPT-6 Sol, where the gap between the two labs' coding agents was already narrow enough to argue about.

What to do about it

• If you build on Replit, open your Agent settings and check which model is actually running your sessions. Don't assume it's the one you picked when you set the project up.
• Treat any platform's default model as a cost decision made on their benchmark, not a verdict on model quality for your use case. Run your own comparison on your actual workload before trusting someone else's internal numbers.
• If you're running model-agnostic automations in n8n or similar tools, audit which provider each AI Agent node is actually calling. Defaults drift quietly, and a provider swap upstream can change your costs without changing your workflow.
• Wait for Anthropic's side of this before drawing conclusions. A one-day walk-back from the CEO who made the original claim, with no comment yet from the other company, is not a settled story.

The short version

Replit's CEO said Anthropic was out, then said he didn't mean it that way, and the people who'd know say the partnership never actually ended. What's real is that a major coding-agent platform quietly moved its default away from Claude, which tells you more about where OpenAI and Anthropic are competing than any one tweet does.