Replit, a platform that lets people build software using AI prompts instead of writing code line by line, has switched the default AI model behind its coding assistant from Anthropic's Claude to OpenAI's models. The company's chief executive confirmed the change over the weekend, framing it as a shift in which model runs automatically when a customer starts a new project — not a removal of Claude from the platform entirely.
For context, Replit is one of a growing crop of "vibe coding" tools that let non-engineers describe what they want in plain English and get working software back. These platforms don't build their own foundation models. Instead, they act as a layer on top of models from OpenAI, Anthropic, Google, and others, routing user requests to whichever model the company has decided works best, is priced most favorably, or has the strongest business relationship at the time.
Until now, Claude was that default for Replit. Anthropic's models have built a reputation in the developer community for being particularly strong at reasoning through code, which is likely why Replit picked Claude as its starting point in the first place. OpenAI's models, meanwhile, have made aggressive moves on both pricing and coding benchmarks over the past year, narrowing whatever edge Claude had.
Customers who prefer Claude can still select it manually inside Replit's settings. What's changed is the out-of-the-box experience — and defaults matter more than most people assume. Industry research on consumer behavior consistently shows that the overwhelming majority of users never touch default settings, which means Replit's choice will quietly determine which company's model actually processes most of the platform's traffic going forward.
This kind of switch isn't unprecedented. As the gap between frontier AI models narrows, app-layer companies like Replit increasingly treat model providers the way airlines treat fuel suppliers — swapping based on cost, performance, and contract terms rather than brand loyalty. Similar behind-the-scenes shuffling has happened at other AI coding and writing tools over the past two years, usually without much public explanation, since the underlying business terms are rarely disclosed.
The move also reflects the broader competitive dynamic between OpenAI and Anthropic, who are fighting for the same pool of enterprise and developer customers. Pricing cuts, context window increases, and coding benchmark wins have become near-monthly events from both companies, and each shift creates pressure on the app-layer companies sitting on top of them to reconsider their defaults.
For small business owners using Replit to build internal tools, customer-facing apps, or quick prototypes, the practical effect may be subtle at first. Code output quality, generation speed, and even monthly costs tied to usage limits can shift when the underlying model changes, since OpenAI and Anthropic price and perform differently depending on the type of coding task. A business running a complex, multi-file project might notice different behavior than one building a simple landing page.
The bigger lesson here is about dependency. Businesses that have built workflows around a specific AI tool's output — the tone of generated code, debugging style, or even typical error patterns — may find those characteristics shift without warning when the provider changes underneath them. This is worth testing directly rather than assuming continuity.
Watch for whether Replit publishes pricing changes tied to this switch, since OpenAI and Anthropic have different cost structures for API usage at scale. Also worth tracking: whether other AI coding platforms — including Cursor, GitHub Copilot, and Bolt — make similar default changes in the coming months, which would suggest this is part of a wider shift in leverage between OpenAI and Anthropic rather than a one-off decision.
If your business relies on Replit for active projects, it's worth running a quick side-by-side test this week — rebuild a recent feature request using both OpenAI and Claude as the selected model — before assuming the new default serves your use case as well as the old one did.