MeanCEO: Tech Startups and Startup Ideas

Replit Makes OpenAI Its Default Over Claude. What Changes for Your Vibe-Coded App (and Your Credits)

TL;DR: Did Replit drop Anthropic's Claude for OpenAI?

Not completely. On 16 October 2026 Replit CEO Amjad Masad said on stage at TechCrunch Disrupt that Replit is "no longer working with Anthropic". Afterwards he clarified that OpenAI is now the default model provider for Replit Agent and that customers can still pick Claude. A person familiar with the deal told TechCrunch the Anthropic partnership is still in place, and Replit's own docs still list seven Claude models. For you, the practical change is the default: if you never touch the model menu, your next build may run on a different model, with different code and a different credit bill.

Most people who vibe-code an app on Replit never open the model menu. They type a prompt, the Agent builds, the credits go down. That's why one sentence from Replit's CEO this week matters more than it sounds.

Replit is moving its default from Anthropic's Claude to OpenAI. The same week, new data from Vercel showed that almost one in three teams switched their main AI lab in a single month, and Accenture capped the Claude budget it funds per employee at $29. The pattern is the same: AI builders are choosing models on price as much as on quality, and the choice is often made for you.

What happened

On Thursday 16 October 2026, Amjad Masad, Replit's CEO and co-founder, told the audience at TechCrunch Disrupt in San Francisco that Replit is no longer working with Anthropic. TechCrunch reported that he gave three reasons: "It was a decision based on both cost, partnership, and performance."

On performance he was blunt: "I think ChatGPT is better now." He said OpenAI's Sol and Astra models "perform better on building vibe-coded applications from scratch".

After the session, Masad walked part of it back. Customers can still choose Claude, but OpenAI is now the default. "Replit remains model agnostic," he told TechCrunch. "OpenAI is now our preferred partner for powering Replit Agent."

The history: Claude models powered Replit before Replit announced a partnership with OpenAI in August, according to TechCrunch. Masad said Replit had recently been one of Anthropic's top five customers.
The dispute: one person familiar with the partnership told TechCrunch it is still in place. TechCrunch also pointed to a late-September Replit blog post that named Claude Fable 5 and Fable 5.1 as models running in Replit Agent's production environment, and to a Replit employee's post on X promoting Claude Opus 5.5.
The update: TechCrunch first published the story at 2:22 pm PT on 16 October and updated it on 17 October "to include more context around Replit's partnerships". The headline changed from "Replit is no longer working with Anthropic" to "Replit makes OpenAI its default over Anthropic".
No comment from the labs: the article includes no statement from Anthropic or OpenAI.

Masad also shared a business number that explains why model costs matter so much to Replit: enterprise was about 2% of revenue last year and about 15% this year, and more than half of revenue now comes from business customers. He didn't give revenue figures.

What actually changes inside Replit

The Replit docs on the Model selector explain how the choice works today. By default, Replit uses "intelligent model routing", shown as Auto: it picks a model per task, "balancing quality and cost", and can switch models as the task changes. Free Mode always uses this routing, and you can't choose a model there.

Builders on Core and Pro can override it. In Power Mode or Max Mode you pick a Primary model, and the choice applies to your next message. Here's the list the docs showed on 18 October:

Agent mode	Models you can pick	Who decides
Free Mode	None: routing only	Replit
Power Mode	GPT-6 Sol, GPT-6 Luna Fast, Claude Sonnet 5, Claude Sonnet 4.6	You, or Auto
Max Mode	GPT-6 Astra, Claude Fable 5.1, Claude Opus 5.5, Claude Opus 4.8, Claude Opus 5, Claude Opus 5 Fast, Kimi K3	You, or Auto

Source: Replit docs, "Model selector", read 18 October 2026. Replit says the list in your own account is the source of truth: model selection is still rolling out, and Fast variants need Pro or Enterprise.

So Claude hasn't disappeared. What has changed is the model Replit reaches for when you don't choose. On Replit's pricing page, Core costs $20 a month ($18 billed yearly) with $20 of credits for the more powerful models, and Pro costs $100 a month ($90 yearly) with $100 of credits. A different default model means a different cost per task inside the same budget. The docs also warn that higher Effort settings can add up to about 2x cost on the hardest tasks.

Why builders are switching: the cost data

Replit isn't alone. On 15 October, Vercel published its AI Gateway Production Index for September, based on traffic through its AI Gateway. The numbers show how fast teams now move between models:

Open-weight models carried 71.5% of tokens and about 23% of spend. In June they took about 5% of spend.
Anthropic's share of spend fell below half for the first time, from 63.7% to 43.9%. Claude Opus 5.5 was still the single biggest model by spend, at 11.03%.
Almost one in two teams changed their most-used model in September, and almost one in three changed their most-used lab, an index record.
DeepSeek-V4.1-Flash carried nearly half of September's tokens but only 2.6% of spend.
The average price per token fell 47% compared with December, while total spend grew about 13 times.

Vercel's caveats: spend is estimated at published list prices, so real bills may differ, and its definition of open-weight is broader than in earlier editions. Still, the direction is clear. Teams keep the expensive frontier models for hard work and send routine tokens to cheap ones.

Big companies feel the same squeeze. 404 Media reported on 15 October that Accenture told many employees it will now fund only $29 a month per person for Claude, starting 1 October. Some had been spending as much as $1,000 a month. According to 24/7 Wall St., anything above the cap must be charged to a project or team budget. An Accenture spokesperson said: "Our approach today is to provide the right tool for the right work."

We covered the price side of this in our AI price war report and in Gemini 4 Argon vs Muse Wonder. For the Claude side of the month, see our Anthropic and Claude news for October 2026.

The European angle: where does your model run?

For founders in Europe, the model is only half the question. The other half is where the data goes. When your builder changes its default provider, the company processing your prompts, and possibly your users' data, can change too.

Two launches this week show European builders treating this as a feature:

Lovable, the Stockholm vibe-coding company, launched EU inference on 12 October for Enterprise plans. Supported model requests, both while building and from Lovable AI features in published apps, go to EU endpoints, and requests that no EU endpoint can serve fail instead of going elsewhere. The catch: Claude is rejected in published apps that use Lovable's built-in AI when the setting is on. Web search, page fetching and browser automation still send data to US services, and the feature covers processing, not storage.
Vercel launched Token Pass in early access for Pro teams on 15 October: $30 a month for $80 of shared credit on open-weight models GLM 5.3 and Kimi K3, with inference in the US and the EU and zero data retention on by default. Unused credit doesn't roll over, and requests simply stop when the credit runs out, with no overage charges.

Builder or service	Default model	Can you choose?	EU inference
Replit Agent	OpenAI (Auto routing)	Yes, on Core and Pro, in Power or Max Mode	Not mentioned on the pages we read
Lovable	Not stated in the EU inference post	Model list depends on your settings	Yes, Enterprise only; Claude blocked in published apps
Vercel Token Pass	GLM 5.3 or Kimi K3	Yes, via your coding agent	Yes, US and EU, zero data retention

Sources: Replit docs and pricing page; Lovable blog, 12 October 2026; Vercel changelog, 15 October 2026.

If your app sends customer data to an AI model, a provider change is also a GDPR question. Check that your data processing agreements and your privacy notice still name the right processors.

What this means for you

Know your model. If you leave Replit on Auto, Replit decides. That's fine for prototypes. For a product you charge money for, open the Model selector and look at what you're actually getting.
Pin the model for serious work. On Core or Pro, pick a primary model in Power or Max Mode for client projects, so a default switch can't quietly change how your code is written halfway through a build.
Test it yourself. Masad says GPT-6 Sol and Astra build vibe-coded apps better; others will disagree. Replit's Compare models beta runs one prompt on up to four models, though each run uses credits. Take one real task from your backlog and judge the output. Our guide on how to evaluate an AI tool gives you a simple scoring method.
Track credits per feature. Monthly totals hide the expensive tasks. Note what each feature cost to build, then decide where you need a frontier model and where a cheap one will do. For a lean stack, read building an MVP on a bootstrap budget.
Don't get locked in. Keep your specs, prompts and code in a repository you control. Vercel's data shows teams switching labs every month; your builder can switch too, and so can you. Our AI tools stack for solo founders lists alternatives.
Budget like Accenture, not like a fan. If a company with about 814,000 employees caps Claude at $29 a head, a solo founder can set a monthly AI ceiling too. Spend the big-model money where it changes the result. For other builders' moves this month, see our Lovable news for October 2026.

Are you getting customers from AI search yet?

More and more traffic is coming from Perplexity, ChatGPT, Grok and other AI tools.

👉 Rank on ChatGPT

FAQ: Replit, OpenAI and Claude

Did Replit stop using Claude?

No. Replit made OpenAI the default model provider for Replit Agent, but CEO Amjad Masad said customers can still choose Claude. On 18 October 2026, Replit's Model selector docs listed seven Claude models across Power and Max Mode.

Why did Replit switch its default to OpenAI?

Masad said it was "a decision based on both cost, partnership, and performance" and that OpenAI's Sol and Astra models build vibe-coded apps from scratch better. Replit announced an OpenAI partnership in August, according to TechCrunch.

Is Replit still partnered with Anthropic?

It's disputed. Masad said on stage that Replit is no longer working with Anthropic, but a person familiar with the partnership told TechCrunch it is still in place. Neither Anthropic nor OpenAI commented in TechCrunch's report.

How do I choose Claude in Replit Agent?

On a Core or Pro plan, open Agent Mode in the prompt box, select Power or Max, and pick a model under Primary model. Free Mode always uses Replit's automatic routing. Replit says model selection is still rolling out, so your list may differ.

Does Replit's default model change affect my costs?

It can. Replit plans include a fixed amount of credits ($20 on Core, $100 on Pro), and different models use those credits at different rates. Higher Effort can add up to about 2x on the hardest tasks, according to Replit's docs.

Which AI app builders offer EU inference?

Lovable offers EU inference on its Enterprise plan since 12 October 2026, with Claude blocked in published apps when it's on. Vercel's Token Pass, in early access for Pro teams, runs GLM 5.3 and Kimi K3 inference in the US and the EU with zero data retention by default.

About the Author

Violetta Bonenkamp, also known as MeanCEO, is an experienced startup founder with an impressive educational background including an MBA and four other higher education degrees. She has over 20 years of work experience across multiple countries, including 5 years as a solopreneur and serial entrepreneur. Throughout her startup experience she has applied for multiple startup grants at the EU level, in the Netherlands and Malta, and her startups received quite a few of those. She’s been living, studying and working in many countries around the globe and her extensive multicultural experience has influenced her immensely.

Violetta Bonenkamp

2026-10-18 05:17 News and Trends