Replit, the platform for building software with AI assistance, has transitioned its default artificial intelligence model provider from Anthropic to OpenAI. Replit CEO Amjad Masad announced the shift, stating that OpenAI's models now offer superior performance for the company's workloads. This change means that when users initiate new projects on Replit, OpenAI's models will be used by default. However, Anthropic's Claude models will still be accessible to users who wish to select them manually.
The decision to switch default providers was based on a combination of factors, including cost, partnership dynamics, and performance benchmarks. Masad indicated that internal and public benchmarks favored OpenAI's models, such as Sol and Astra, for tasks like building applications from natural language instructions. Replit's strategy involves using different AI models for specific tasks to achieve optimal quality at the lowest possible cost, rather than relying on a single provider. This approach positions Replit as a model-agnostic platform, allowing flexibility in adapting to the rapidly evolving AI landscape.
The move highlights the dynamic nature of partnerships within the artificial intelligence sector, where companies frequently reassess their AI model suppliers based on performance, pricing, and strategic alignment. Replit's enterprise business has seen significant growth, with business-to-business clients now accounting for over 50% of its revenue and enterprise contracts increasing substantially. This commercial expansion may have influenced the decision to prioritize cost-effectiveness and performance in its AI model choices.
While Masad's public statements suggested a complete cessation of Replit's work with Anthropic, some individuals familiar with the partnership have disputed this claim. These sources indicated that the partnership with Anthropic may still be active, and Replit's own documentation from September reportedly continued to list Claude models in its production environment. Replit has not provided further clarification on the precise status of its relationship with Anthropic.
Replit's platform enables users, including those without extensive coding experience, to describe desired software functionalities in plain English, with AI models generating the code. This "vibe coding" approach relies heavily on the underlying AI models for code generation, application scaffolding, and agent-driven workflows. The choice of AI model provider directly impacts Replit's platform capabilities, cost structure, and competitive positioning within the AI-assisted coding market.