Replit Just Made OpenAI the Default. Claude Is Still There — If You Go Looking.

TL;DR: Replit has made OpenAI its default model provider, replacing Anthropic. Claude hasn’t been removed — CEO Amjad Masad says “Replit remains model agnostic” and users can still choose it. But the quiet part is the important part: most people never change a default, and Replit’s routing has been picking models for you since August anyway. If you came for Replit Claude access specifically, this is the week to check which mode you’re on.

What happened

At TechCrunch Disrupt on Thursday, Replit CEO and co-founder Amjad Masad said the company was no longer working with Anthropic. TechCrunch reported it on 16 October, then updated the piece the following day with added context and a changed headline: Replit has made OpenAI its default over Anthropic.

Onstage, Masad gave three reasons: “It was a decision based on both cost, partnership, and performance.” On performance he cited an internal benchmark alongside a public one and said OpenAI’s Sol and Astra do better at building vibe-coded apps from scratch. His line was blunt: “I think ChatGPT is better now.”

In a later statement to TechCrunch he put it more precisely: “Replit remains model agnostic”, and “OpenAI is now our preferred partner for powering Replit Agent, based on performance, economics.” Users, he said, “can choose Anthropic’s Claude or other supported models.”

So the accurate version is narrower than the first headline: a default changed. Replit Claude access was not removed.

Replit Claude access, in Replit’s own documentation

Worth confirming against the product rather than the stage, because this is where it actually matters.

Replit’s changelog, dated 2 October 2026, announces new OpenAI and Anthropic models, and says plainly: “You can now choose GPT-6.1 Sol and Claude Sonnet 5.5 when building with Agent.” It adds that “Sonnet 5.5 is available in Power Mode, and GPT-6.1 Sol is available in Max Mode.”

Replit’s engineering post of 29 September, written by five of its researchers, describes Agent being evaluated “with Astra as the core loop”, where “GPT-6 Astra hands routine implementation to less costly subagents”. Its first table compares Replit Agent production delegation across Fable 5, Fable 5.1 and GPT-6 Astra — Fable being Anthropic’s.

So Replit Claude access is documented, recent, and — per Masad — continuing. TechCrunch also reported an unnamed source saying the Anthropic partnership was still in place, which sits awkwardly beside the CEO’s account but no longer changes the practical answer.

The quiet part Replit Claude users should notice

Here is the sentence in TechCrunch’s updated piece that deserves more attention than the headline: most customers stick with the default.

That is why a default change is not a small thing dressed up as a big one. It’s the opposite. Nobody has to be persuaded, nobody gets an email, no setting breaks. The models simply change underneath the people who never open the menu — which, on any platform, is most of them. Replit Claude access surviving matters a great deal less if nobody goes and selects it.

And it compounds with something we covered in August: since 26 August, Replit’s routing has picked the model for you. We wrote then that this changed what your money buys. Put the two together and the honest position for a typical user is: you were already not choosing, and now what you’re not choosing has changed. Replit Claude access is a menu item, not a default.

What this does and doesn’t mean for you

Question	Answer
Is Replit Claude access gone?	No. The 2 October changelog offers Claude Sonnet 5.5 in Power Mode, and Masad says users can still choose it
What actually changed?	The default. OpenAI is now the preferred partner powering Replit Agent
Will my builds get worse?	No evidence either way. Masad’s performance claim rests on Replit’s own internal benchmark
Will it get cheaper for me?	Cost was a stated reason for Replit. Whether a supplier saving reaches your bill is a separate decision nobody has announced
Do I need to do anything?	Only if you specifically chose Replit for Claude

What to actually do

1. If you never picked a model, you have a decision you didn’t know you had. Open the model settings once and look at what’s selected — that is where Replit Claude access lives. That’s the whole job.
2. If you’re on a paid tier, check which mode you’re in. Power Mode and Max Mode currently carry different models, so this is the one place Replit Claude access is visible to you.
3. Re-run a prompt you ran last month and compare the output. When routing and defaults shift underneath you, your own work is the only early warning you get.
4. Bookmark Replit’s changelog, not its blog. Product facts land there first, and if Replit Claude access ever does go, that’s where it will go first.
5. Don’t pick a builder on which model it uses this quarter. Defaults rotate. Pick on whether you can export your work.

Who should care (and who shouldn’t)

• Building on Replit and never touched the model settings: this is aimed at you. Step 1 takes a minute and tells you whether Replit Claude is running your builds.
• Specifically chose Replit for Replit Claude access: check your mode this week. Nothing has been taken away, but the default is no longer on your side.
• Choosing between builders: don’t let this decide it. Every one of these platforms has changed models this year.
• On Lovable, Bolt or Figma Make: nothing here touches you, though the lesson travels — the model under your builder is not a fixed thing.
• Watching the industry: a platform can now swap its frontier supplier and tell users honestly that nothing was removed. That’s new, and it says a lot about where the leverage has moved.

Our take

The revised headline is the fair one, and the first one wasn’t. “No longer working with Anthropic” and “made OpenAI the default” are very different claims, and the gap between them is exactly the kind of thing that gets repeated wrongly for months. Credit to TechCrunch for updating it; the lesson for readers is to check the changelog rather than the keynote.

But we’d push back on the idea that a default change is minor. It is the single most powerful lever a platform has, precisely because it requires nothing of the user. Masad is right that Replit remains model agnostic in the sense that matters legally and technically. He is also describing a change that will move most of his users off Replit Claude and onto different models without any of them deciding to.

On the performance claim, treat it as what it is. The benchmark is Replit’s own, and the CEO is describing the models he has just chosen. That is not dishonest, but it isn’t independent either, and we’d want someone outside Replit testing the same thing before repeating it as fact.

The broader shift is the real story. A year ago, which frontier model a builder ran was its headline feature. Now a CEO can change supplier onstage and the honest answer to “what does that mean for users?” is probably nothing you’ll notice — the same shift we saw when Meta put its best model behind a subscription and nobody’s workflow moved. Models have become components. Good for prices; bad for anyone choosing a tool on the strength of the logo underneath it. Replit has also spent this quarter buying its way into analytics, which tells you where it thinks the value actually sits.

Not sure which builder you should be on in the first place? Take the 60-second Vibe Coding Tool Finder quiz →

FAQ