For months, Replit and Anthropic looked like the poster pair of the AI coding boom. Then Replit’s CEO stood on stage at TechCrunch Disrupt and said the quiet part out loud: Replit is no longer working with Anthropic. OpenAI’s ChatGPT is now the default model provider for the platform.
The decision touches cost, performance, and partnership politics at once. If you build apps with AI tools, the model behind your IDE matters more than you think. Here is what changed, why it changed, and what it signals for the vibe coding market.
What Replit’s CEO actually said
Amjad Masad, Replit’s CEO and co-founder, made the announcement during a session at TechCrunch Disrupt. His reasoning covered three fronts:
- Cost. Running a coding platform where users build and deploy apps from the browser means model bills stack up fast. A cheaper default changes the unit economics of the whole product.
- Performance. “I think ChatGPT is better now,” Masad said. Replit uses an internal benchmark to assess model performance, and it also publishes a public one.
- Partnership. Replit announced its OpenAI partnership back in August, but at the time nobody said Anthropic was being pushed aside. Masad framed the final call as “a decision based on both cost, partnership, and performance”.
The performance claim came with specifics. Masad said that on Replit’s benchmarks, Sol and Astra, the platform’s own agent products, “consistently perform better on building vibe-coded applications from scratch” when running on OpenAI models. For a product where the whole point is generating working apps from plain English prompts, that is the metric that counts.
What changes for Replit users
After the talk, Masad clarified the details, and this is the part users should read twice:
- OpenAI’s ChatGPT is now the default model on the platform.
- Anthropic models are not gone. Claude is still available for customers who want to pick it.
- Most customers never touch the model selector. They stay on whatever the platform recommends.
That last point is where the real impact sits. Defaults quietly decide where the money goes. If the majority of Replit users build on the default, Replit’s spend with Anthropic drops significantly, and Anthropic loses a flagship distribution partner.
Replit sells a cloud-based development environment where users build and deploy apps directly in the browser. It has been a favourite of startup founders and first-time developers, and it has leaned hard into the “vibe coding” wave where prompts replace boilerplate. The model that powers that experience shapes everything from code quality to how often the agent breaks your build.
For users on paid plans, expect the practical differences to show up in small places: how often the agent understands a half-written instruction, how clean the generated code looks, and how many tokens a typical build burns. Those details add up when you ship every week.
Why model defaults matter in the AI coding market
This is not just one platform swapping engines. It is a sign of how fast the AI coding market is rearranging.
- Defaults are market share. Most developers use what is in front of them. Whatever model sits in the default slot wins the bulk of the traffic, the training feedback, and the revenue.
- Benchmarks now decide partnerships. Replit did not switch on vibes alone. It ran internal tests and published public benchmarks. Expect more AI product companies to treat model choice as a procurement decision backed by data.
- Everyone is competing for the same layer. OpenAI, Anthropic, and Google are all pushing coding models and agentic tools. Meanwhile Replit’s CEO has said he sees the future of work as the “self-driving company”, where agents handle whole workflows. Owning the default model in that world is worth fighting for.
- Switching costs are dropping. Replit kept Claude available. When platforms support multiple providers, users can move between models, and vendors have to keep earning their spot.
For Anthropic, the sting is real. Claude built a strong reputation in coding tasks, and partnerships like the one with Replit put it in front of millions of builders. Losing default status does not end that reach, but it puts a dent in the growth story.
The road to this switch
The breakup did not come out of nowhere. Replit announced its partnership with OpenAI in August, positioning ChatGPT inside the platform and rolling out features like a free mode for new builders. At the time, Anthropic models were still the engine behind most Replit sessions, and nobody flagged a change.
Through the autumn, the signals piled up. Replit shipped agent updates tuned around OpenAI models. Masad started talking publicly about the “self-driving company”, a workplace where AI agents handle whole workflows instead of single tasks. In that vision, the model partner is not a vendor line item. It is the engine of the product.
When Masad finally said Replit was “no longer working with Anthropic” at TechCrunch Disrupt, the clarification came fast: Claude stays on the platform as an option, but OpenAI now sets the default. The distinction matters commercially. In most software products, the default wins 70 to 90 percent of usage, so a default swap of this size moves real money and real workloads between two of the biggest names in AI.
It also puts pressure on other AI coding platforms. Cursor, Windsurf, and the IDE plugins from all three major model providers are now competing on the same axis: which model builds working apps fastest, and which partnership keeps prices sane at scale.
What Indian developers should take away
India has one of the largest developer communities using AI coding tools, and many startups here build on platforms like Replit to ship fast. A few practical lessons from this switch:
- Do not hardcode your workflow to one model. Today’s default can change with one board meeting. Keep prompts, tests, and code review habits portable.
- Run your own benchmark. Replit picked its default based on internal tests. Test your real tasks, not just demo prompts, before committing a project to one provider.
- Watch your plan pricing. Platform costs track model costs. When a platform renegotiates its model supply, pricing and plan limits often move soon after.
- Keep humans in review. Whether the model is ChatGPT or Claude, generated code still needs testing and security review before it touches production.
There is also a hiring angle. As platforms standardise on particular models, job descriptions increasingly mention specific toolchains. Knowing how to drive both OpenAI and Anthropic models keeps you useful no matter which vendor a company picks next quarter.
FAQ
Why did Replit stop working with Anthropic?
Replit’s CEO cited cost, performance, and partnership as the three reasons. On Replit’s benchmarks, OpenAI models performed better at building apps from scratch, and the OpenAI partnership announced in August gave Replit stronger commercial terms.
Can Replit users still use Claude models?
Yes. Anthropic models remain available on the platform. Only the default has changed to OpenAI’s ChatGPT, so users who prefer Claude can still select it.
What are Sol and Astra on Replit?
They are Replit’s agentic building products that create vibe-coded applications from prompts. Masad said they “consistently perform better” when powered by OpenAI models on both internal and public benchmarks.
Does this mean Claude is worse at coding now?
Not necessarily. Model quality is measured against specific workloads, and Replit’s tests focus on building apps from scratch in a browser environment. The switch shows that model leadership is being decided task by task, not once for all use cases.
What should I do if my Replit projects depend on Claude?
Nothing breaks. Claude remains available on the platform, and existing projects keep their model choice. Just open your account settings and confirm the model per project if you want to avoid being moved to defaults.
