Replit makes OpenAI default model, ending Anthropic exclusivity

Replit has officially shifted its primary AI partnership from Anthropic to OpenAI, a decision co-founder and CEO Amjad Masad announced on stage at TechCrunch Disrupt. According to TechCrunch, Masad stated that while Replit is no longer working exclusively with Anthropic, the company has not completely cut ties; instead, OpenAI’s ChatGPT models are now the default for the platform’s agent capabilities. This marks a significant pivot for the AI coding platform, which previously relied heavily on Anthropic’s Claude models to power its core development environment.

Masad explained that the decision was driven by a combination of cost, partnership dynamics, and performance metrics. "I think ChatGPT is better now," Masad said, noting that Replit uses internal benchmarks to assess performance. He added that their public benchmarks consistently show that Replit’s internal models, Sol and Astra, perform better when building vibe-coded applications from scratch, but the underlying foundation models provided by OpenAI offered a superior balance of economics and capability for their specific use case. "It was a decision based on both cost, partnership, and performance. And for all these reasons, we just felt like they were a better partner for now," he said.

The shift in default models

The practical implication of this announcement is that new and existing users will encounter OpenAI models by default when using Replit Agent. Masad clarified that Replit remains "model agnostic," meaning customers can still manually select Anthropic’s Claude or other supported models if they prefer. However, industry observers note that most users tend to stick with the default option rather than navigating model selection menus, meaning this shift could significantly reduce Replit’s volume of API calls to Anthropic.

Prior to this change, Replit was reportedly a top-five customer for Anthropic, a relationship that helped drive significant revenue growth for the AI lab. The partnership had been a cornerstone of Replit’s strategy, allowing developers to build and deploy applications directly in the browser using Claude’s capabilities. Now, with OpenAI as the preferred partner for powering Replit Agent, the balance of power in the AI coding tool market shifts toward the company behind ChatGPT and GPT-4o.

Conflicting signals and recent history

Despite Masad’s on-stage declaration, insiders familiar with the partnership have pushed back, suggesting the relationship with Anthropic is still technically in place. TechCrunch reported that one person familiar with the deal, who requested anonymity, said the partnership was not terminated. Evidence supports this nuance: as recently as late September, Replit’s blog posts cited Anthropic’s Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent’s production environment. Additionally, Replit product employees have recently promoted the capabilities of Anthropic’s flagship Opus 5.5 model on social media, indicating continued operational integration.

This suggests that while OpenAI is now the strategic and default partner, Anthropic remains a supported backend. The distinction between being a "preferred partner" and an "exclusive partner" is crucial here. Replit is not removing Anthropic from the stack but is demoting it from the primary driver of its agent experience to an alternative option.

Enterprise growth and safety focus

During the same Disrupt appearance, Masad shared insights into Replit’s evolving revenue model, highlighting a rapid shift toward enterprise customers. "Enterprise was like 2% of revenue last year," Masad said. "This year, I would say like 15% of our revenue is enterprise contracts, and more than 50% of our revenue is B2B." He expects more than half of Replit’s revenue to come from enterprise and government customers by 2027, signaling a move away from purely consumer-focused subscriptions.

Masad also emphasized the technical infrastructure required for safe AI deployment, drawing an analogy between human work environments and AI agent habitats. "Agents need a habitat for them to exist and do good work," he said. This habitat includes "good orchestration, good error recovery, good safety and security and sandboxing, good networking, good identity and authorization." He argued that once these foundational elements are solved, AI can be deployed safely, a perspective that likely informs Replit’s choice of partners who can support such robust infrastructure requirements.

What this means for developers

For developers using Replit, the immediate change is a different default model experience. If you rely on specific Claude behaviors or prompt engineering techniques tailored to Anthropic’s models, you may need to adjust your workflows or explicitly select Claude in the model picker. Since Replit is model-agnostic, there is no loss of functionality, but the default shift suggests that OpenAI models are currently optimized for Replit’s specific agent tasks, such as code generation, debugging, and application deployment.

Developers should monitor the performance differences between the default OpenAI models and the available Anthropic alternatives. While Masad claims OpenAI is "better now" for their use case, individual experiences may vary depending on the complexity of the project. The availability of multiple models also provides a hedge against vendor lock-in, allowing teams to switch back and forth based on cost or performance needs.

What to watch

• Anthropic’s response: Whether Anthropic publicly addresses the shift or announces new terms for Replit will indicate the true state of the relationship.
• Performance benchmarks: Replit’s public benchmarks comparing Sol, Astra, and the underlying foundation models will provide concrete data on the performance claims.
• Enterprise adoption: The pace at which Replit converts its B2B revenue to enterprise contracts will test the viability of its new partnership structure with OpenAI.
• Model availability: Any changes to the list of supported models or pricing tiers for Anthropic vs. OpenAI models on the Replit platform.