Last updated: October 18, 2026, 10:30 AM IST
Replit just made one of the boldest moves in the AI coding race. The company's CEO walked onto the stage at TechCrunch Disrupt and told the world that Replit is no longer working with Anthropic — and that OpenAI is now its preferred partner for powering Replit Agent.
That single statement sent a small shockwave through the developer world. For a long time, Replit was one of Anthropic's biggest customers, and Claude was the engine behind much of what made Replit Agent feel so good to use. Now, ChatGPT models — specifically OpenAI's Sol and Astra — sit in the default slot. But the story didn't end on that stage. Two days later, the CEO was already clarifying his words on X, and a source told TechCrunch the Anthropic partnership may not actually be dead. Here's the full picture of what changed, what didn't, and what it means if you code with Replit.
What exactly did Replit announce?
On October 15, 2026, Replit CEO and co-founder Amjad Masad spoke onstage at TechCrunch Disrupt. He said Replit is "no longer working with Anthropic" and named OpenAI as the company's preferred partner for powering Replit Agent, the company's AI coding assistant that builds apps from natural-language prompts.
The practical change is this: when you open Replit Agent today, the model running behind the scenes by default is OpenAI's ChatGPT family — Sol and Astra models. If you preferred Claude, you can still pick it by hand. The default simply switched.
This didn't come out of nowhere. Replit and OpenAI announced a partnership back in August 2026, so the two companies have been working together for a couple of months. The October announcement was the moment Replit made the pecking order official: OpenAI first, Anthropic second. As Masad put it in his Disrupt interview, he now believes "ChatGPT is better" for what Replit needs.
Can you still use Claude in Replit?
Yes — and this is the question most Replit users are actually asking. Claude has not been removed from the platform. You can still manually select Anthropic's Claude models inside Replit Agent. What's changed is only the default: new sessions start on OpenAI's models unless you switch.
Claude is still deeply woven into Replit's recent history. Production records from late September 2026 showed Claude Fable 5 and Fable 5.1 running inside Replit Agent, and the October 2 changelog added both GPT-6.1 Sol (in Max Mode) and Claude Sonnet 5.5 (in Power Mode). So the transition is gradual, not an overnight switch. If your workflow is built around Claude, nothing has been ripped away — you just need to pick it explicitly for now.
Why OpenAI over Claude? Three reasons Masad gave
Masad pointed to three things: cost, performance benchmarks, and the strength of the partnership itself.
1. Cost. Running an agentic coding product is expensive — every prompt a user types burns through tokens on Replit's bill. On X, on October 17, Masad clarified that "OpenAI models are generally more affordable." For a company where AI compute is the single biggest cost, that matters a lot.
2. Benchmarks. Replit built its own internal benchmark called VBench, and it also points to its public benchmark results. According to the company, Sol and Astra "perform better on building vibe-coded applications from scratch" — meaning when you describe an app in plain English and the agent builds the whole thing, OpenAI's models come out on top in Replit's own testing.
3. The partnership. Replit and OpenAI formalized their relationship in August 2026, and Masad described OpenAI as the stronger, more committed partner right now. In the AI model wars, the big labs are competing hard to power the most popular developer tools — and OpenAI clearly won this round for Replit.
- What's now default: OpenAI's ChatGPT (Sol and Astra models) in Replit Agent
- Still available: Anthropic's Claude, selectable manually — Fable 5/5.1 and Sonnet 5.5 have all run on the platform
- Why the switch: lower cost, better VBench results on vibe-coded apps, stronger partnership
- When it started: announced October 15, 2026 at TechCrunch Disrupt; rolling out gradually
The twist: is the Anthropic partnership really over?
Here's where the story gets messy. Masad's onstage line was blunt — Replit is "no longer working with Anthropic." But two days later, he was on X softening the message, talking about OpenAI as the "preferred partner" and Claude as still selectable. And an unnamed person familiar with the relationship told reporters the Anthropic partnership is actually still in place.
There's more smoke here: shortly before the announcement, Replit's own site and blog were still promoting Anthropic products. Companies don't usually market a partner's products days before dumping them — which suggests the "breakup" may be more of a re-ranking than a clean cut.
So be careful how you read Masad's quote. It's his stated position onstage, not a confirmed fact about the business relationship. The safest way to describe what's happening: OpenAI now runs the default experience in Replit Agent, Claude stays available as an option, and the two companies are telling slightly different stories about whether the partnership itself survives. Reality is probably somewhere in between — a partnership that's cooling, not one that's been switched off.
What this means if you build with Replit
For most Replit users, day-to-day life changes very little — but there are a few things worth knowing.
Your default model changed. New Replit Agent sessions now run on Sol and Astra. If you liked Claude's coding style, check the model picker and switch back. Don't assume the output will read exactly the same; every model has its own quirks with code structure and explanations.
Watch your usage costs. One reason Replit made this move is that OpenAI's models are cheaper to run, which is good news for Replit's margins — and potentially for how much agent time your plan buys you. No price changes have been announced, but cheaper underlying models give the company room to be generous later.
Expect more OpenAI-flavored features. When a platform names a "preferred partner," that partner's newest models usually land there first. Replit users will likely see OpenAI's latest capabilities show up in the Agent before they arrive anywhere else.
This is bigger than Replit. The AI coding assistant space is consolidating fast. Tools like Cursor lean on their own model mixes, while Replit has now publicly picked a side. For developers, the practical lesson is simple: don't marry one model. Learn to switch between them, because the "best" coding model changes every few months.
Replit's bigger bet behind the switch
The model swap isn't just about which chatbot writes better code — it's about where Replit's money comes from. According to coverage of Masad's Disrupt remarks, enterprise contracts have grown from roughly 2% of Replit's revenue to about 15% this year, and more than half of the company's revenue now comes from B2B customers.
That changes the math completely. When your revenue comes from businesses building real products, the cost of every agent run matters enormously. A cheaper model that scores well on Replit's own benchmarks isn't a small optimization — it's the difference between a healthy business and one that bleeds money on compute. OpenAI winning the default slot is, at its core, a business decision dressed up as a technology announcement.
It's also worth remembering how far this relationship has swung. Until recently, Replit was one of Anthropic's top-five customers, and the Claude partnership was credited with driving serious revenue growth for Replit. Model loyalty in this industry lasts exactly as long as the benchmarks and the bills say it should. OpenAI's ChatGPT lineup is having a strong year across the board, and Replit is just the latest platform to bet on it.
Frequently Asked Questions
Can you still use Claude in Replit?
Yes. Claude hasn't been removed — it's just no longer the default. You can still select Anthropic's Claude models manually inside Replit Agent whenever you start a session.
What are Sol and Astra, the models Replit now uses by default?
They're OpenAI's ChatGPT-family models that now power Replit Agent out of the box. Replit says they scored better on its internal VBench benchmark for building complete applications from plain-English descriptions.
Why did Replit switch from Claude to OpenAI?
CEO Amjad Masad gave three reasons: OpenAI's models are generally more affordable to run, they performed better on Replit's benchmarks for vibe-coded apps, and OpenAI is currently the stronger, more committed partner.
Did Replit completely end its partnership with Anthropic?
That's disputed. Masad said onstage that Replit is "no longer working with Anthropic," but he later softened that on X, and a source told TechCrunch the partnership remains in place. Treat the breakup claim as his stated position, not confirmed fact.
Will this change my Replit plan or pricing?
No pricing changes have been announced. The switch is about which AI model powers Replit Agent by default. If anything, cheaper underlying models could give Replit room to offer more agent usage per plan in the future — but that's speculation, not a promise.
Sources
- TechCrunch — "Replit makes OpenAI its default over Anthropic, CEO says"
- TokenPost — "Replit CEO Says Company Has Ended Its Work With Anthropic"
- WebProNews — "Replit Names OpenAI Its Preferred Partner, Making ChatGPT the Default Over Anthropic's Claude"
My Personal Take
Honestly, I think this is less dramatic than the headlines make it sound — and more interesting. Masad's "no longer working with Anthropic" line was great theater for a Disrupt stage, but the follow-up told the real story: Claude is still there in the model picker, the partnership may not even be dead, and this is really about unit economics. Replit burns an enormous amount of money on tokens every day. If OpenAI gives them similar quality at a lower price, switching the default is just good business, not betrayal.
My advice for anyone building on Replit: spend ten minutes comparing Sol/Astra against Claude on your own project before you form an opinion. Benchmarks like VBench are built by the company selling you the switch — your codebase is the only benchmark that actually matters. And whichever model you pick, keep both in your toolkit. In the AI coding world, today's default is tomorrow's backup plan.
0 Comments