Replit Makes OpenAI Its Default AI Partner

Replit has made OpenAI its preferred partner for powering its AI coding agent, replacing Anthropic as the platform’s default model provider, according to co-founder and CEO Amjad Masad. The decision reflects differences in model performance, operating costs and the strength of the companies’ commercial partnerships.

Masad said Replit had stopped working with Anthropic for several reasons, including cost and performance. However, he later clarified that Anthropic’s models remain available to users, meaning the change concerns Replit’s default provider rather than a complete removal of Claude from the platform.

The shift follows Replit’s announcement of a partnership with OpenAI in August. Before that deal, the company relied heavily on Anthropic’s Claude models to power its AI-driven software development tools.

Why Replit Chose OpenAI Over Anthropic

Masad said Replit evaluates AI models using an internal benchmark and also publishes a public benchmark to compare their performance.

He argued that OpenAI’s models currently perform better at building applications from scratch, particularly in workflows associated with so-called vibe coding, where users describe what they want to build in natural language and let AI generate much of the implementation.

“I think ChatGPT is better now,” Masad said, explaining that the decision reflected performance, cost and the broader partnership.

Replit’s preference for OpenAI does not necessarily mean Anthropic’s models have been removed from its infrastructure. Following his comments at Disrupt, Masad clarified that customers can still select Claude and other supported models.

“Replit remains model agnostic,” Masad said. “Our users can choose Anthropic’s Claude or other supported models…OpenAI is now our preferred partner for powering Replit Agent, based on performance, economics, and the strength of our partnership.”

The distinction matters because many users may continue with the platform’s default model rather than manually selecting an alternative. Making OpenAI the default could therefore shift a substantial share of Replit’s model usage and spending towards the company, even while Claude remains accessible.

Anthropic Partnership Questions Remain

The announcement has also raised questions about the status of Replit’s relationship with Anthropic.

Masad said onstage that Replit had until recently been one of Anthropic’s five largest customers. However, people familiar with the relationship disputed the suggestion that the partnership had ended, according to TechCrunch.

There are also signs that Anthropic models remained part of Replit’s product environment in the weeks before the announcement.

A Replit blog post published in late September referenced Anthropic’s Claude Fable 5 and Claude Fable 5.1 as models operating in the Replit Agent production environment. A Replit product employee also promoted Anthropic’s Opus 5.5 on social media around the same period.

These references do not establish the current extent of the commercial relationship, but they suggest that the transition may be more nuanced than a complete break.

For now, Replit’s stated position is that it remains model-agnostic while giving OpenAI preferred-partner status.

Replit’s Enterprise Business Is Growing

Masad also discussed Replit’s changing revenue mix, highlighting the growing contribution of enterprise customers.

According to the CEO, enterprise contracts accounted for about 2% of Replit’s revenue in 2025. He said the share had increased to approximately 15% in 2026, while business-to-business customers now generate more than half of the company’s revenue.

Replit has not disclosed its annual recurring revenue (ARR), so the figures provide a picture of its customer mix rather than its overall revenue scale.

Masad said he expects enterprise and government customers to account for more than half of Replit’s revenue in 2027.

That would mark a significant development for a platform popular with individual developers, startup founders and small teams. Replit’s browser-based environment allows users to build, test and deploy applications without setting up a traditional local development environment.

Its AI tools have broadened that proposition by allowing people to create applications through natural-language instructions, potentially making software development more accessible to users with limited coding experience.

A larger enterprise business could also mean greater demand for security controls, administrative tools, access management and predictable costs. These requirements are particularly relevant when organisations use AI agents to write and execute code or interact with business systems.

AI Coding Agents Need Stronger Safety Controls

Masad also addressed the infrastructure needed to deploy AI agents safely.

He argued that agents require an environment designed to support reliable work, including orchestration, error recovery, security, sandboxing, networking, identity and authorisation.

The comparison reflects a growing concern in AI development: generating code is only one part of building a reliable AI-powered software development platform. Systems must also control what agents can access, recover from failures and prevent unintended actions.

Sandboxing is especially important for coding agents because they may execute generated code, install dependencies or interact with files and services. Appropriate isolation can reduce the risk that faulty or malicious code affects other systems.

As Replit expands its enterprise business, these controls could become increasingly important to customers that need to meet internal security requirements and manage access across teams.

What the Shift Means for AI Coding Platforms

Replit’s decision highlights the growing competition among AI model providers to become the underlying technology behind software development platforms.

For coding companies, choosing a model provider involves more than comparing benchmark scores. Performance, inference costs, reliability, commercial terms and access to new capabilities can all influence the economics of delivering AI-powered tools.

Making one provider the default can also shape user behaviour and determine how a platform distributes its model usage. At the same time, retaining access to competing models gives users flexibility and reduces dependence on a single provider.

Replit’s approach leaves room for both strategies: OpenAI receives preferred status, while users can continue choosing Anthropic and other supported models.

The longer-term outcome will depend on how the models perform in real-world development tasks, how their costs change and whether Replit can sustain growth among enterprise customers.

For now, the company’s decision marks a change in its AI partnership strategy, but not the end of model choice on its platform.