Replit Switches Default AI to OpenAI, Citing Performance and Cost

Overview

Replit has officially shifted its default AI model provider from Anthropic to OpenAI, a strategic move that underscores the intensifying competition between the two leading AI developers in the coding assistant space. Announced during a keynote at TechCrunch Disrupt on October 16, 2026, the decision marks a significant pivot for the cloud-based development platform, which has long relied on Anthropic’s Claude models to power its core agent experiences.

Amjad Masad, Replit’s CEO and co-founder, revealed that the switch was driven by a combination of performance metrics, economic considerations, and partnership dynamics. While the move positions OpenAI’s ChatGPT as the primary engine for Replit’s vibe-coding features, Masad clarified that the platform remains model-agnostic, allowing users to still access Anthropic’s offerings if they choose.

"We publish a public benchmark as well, and consistently, Sol and Astra perform better on building vibe-coded applications from scratch. It was a decision based on both cost, partnership, and performance." — Amjad Masad, Replit CEO

This shift comes just months after Replit announced a broader partnership with OpenAI in August 2026. However, at the time, the company did not indicate that its longstanding relationship with Anthropic was ending. The transition appears to be a recalibration of priorities rather than a complete severing of ties, though the change in default settings is likely to have substantial financial implications for both AI providers.

Performance and Economic Drivers

The primary catalyst for Replit’s decision appears to be the comparative performance of OpenAI’s models versus Anthropic’s when handling complex, generative coding tasks. Masad highlighted that Replit utilizes both internal benchmarks and public evaluations to assess how well different models handle the specific demands of its platform, particularly the creation of applications from scratch.

According to Masad, OpenAI’s recent iterations, referred to by the codenames Sol and Astra, have demonstrated superior capabilities in these benchmarks. The term "vibe-coding" has become central to Replit’s value proposition, referring to the ability of users to describe an application concept in natural language and have the AI generate a fully functional, deployable product. In this high-stakes environment, where the quality of the generated code directly impacts user retention and satisfaction, incremental performance gains are critical.

Cost efficiency also played a decisive role in the executive decision. As enterprise demand for AI-driven development tools grows, the margin costs associated with API calls to large language models can significantly impact a company’s bottom line. By switching the default to OpenAI, Replit anticipates reducing its overall spend, assuming the pricing structures and performance ratios favor OpenAI for their specific workload volumes.

The partnership with OpenAI, initially signaled in August, is now bearing fruit in the form of a deeper integration. Masad described the relationship as robust, noting that the strength of the partnership contributed to the choice alongside raw technical metrics. This alignment suggests that OpenAI may offer Replit more favorable terms or closer collaborative opportunities compared to other providers.

Impact on Anthropic and User Choice

Despite the headline-grabbing nature of the switch, Replit is not entirely abandoning Anthropic. Masad emphasized that customers retain the ability to select Anthropic’s Claude models or other supported options within the platform. However, the reality of user behavior suggests that the default setting carries immense weight; the vast majority of users tend to stick with the pre-selected model rather than navigating to advanced settings to switch providers.

Prior to this change, Replit had been one of Anthropic’s top-five customers, a relationship that contributed significantly to Anthropic’s revenue growth during its rapid expansion phase. The move by Replit to elevate OpenAI could therefore signal a measurable dip in Anthropic’s enterprise footprint, at least from one of its most visible partnerships.

Insiders familiar with the Anthropic-Replit relationship have expressed skepticism about the finality of the split. One source, speaking on condition of anonymity, stated that the partnership remains active. This sentiment is supported by evidence from Replit’s own marketing channels. As recently as late September, Replit published blog posts highlighting the use of Anthropic’s Claude Fable 5 and Claude Fable 5.1 in production environments. Additionally, Replit product staff have publicly promoted the capabilities of Anthropic’s flagship Opus 5.5 model on social media.

This dichotomy between Masad’s public statements and the ongoing promotional activity suggests a nuanced transition rather than an abrupt cutoff. Replit may be balancing its commitments, using OpenAI for its most prominent agent features while maintaining Anthropic as a viable alternative for power users who prefer its specific strengths in reasoning and safety.

Enterprise Growth and Future Outlook

Beyond the AI provider controversy, Masad used the TechCrunch Disrupt stage to provide a broader picture of Replit’s commercial trajectory. He revealed that enterprise revenue has grown from approximately 2% of total income last year to an estimated 15% this year. Furthermore, more than half of Replit’s current revenue is derived from B2B contracts, indicating a strong shift toward corporate adoption.

Looking ahead to 2027, Masad projects that enterprise and government customers will account for more than half of the company’s revenue. This focus on larger organizational clients aligns with the broader trend in the AI industry, where individual developer tools are increasingly being scaled into institutional workflows with stricter security and compliance requirements.

"In the same way that us humans need a certain environment to work... agents need a habitat for them to exist and do good work."

Masad also touched on the philosophical and technical necessities of deploying AI agents at scale. He argued that for AI to be safe and effective, it requires a robust "habitat" characterized by good orchestration, error recovery, sandboxing, and identity management. This perspective reinforces Replit’s value proposition as a secure, contained environment for AI-assisted development, distinguishing it from standalone model providers.

As the AI coding landscape continues to evolve, Replit’s decision to favor OpenAI while maintaining Anthropic access reflects a pragmatic approach to product development. The company is prioritizing the tools that currently offer the best balance of performance and cost for its mainstream user base, while keeping doors open for the diverse needs of its growing enterprise sector. The coming months will likely reveal whether this strategy solidifies Replit’s position as a leader in the vibe-coding era or if the nuanced partnership dynamics with Anthropic will eventually re-emerge as a competitive differentiator.