Replit, the cloud-based coding platform popular with startup founders, has shifted its primary artificial intelligence provider from Anthropic to OpenAI. CEO and co-founder Amjad Masad announced the change on Thursday during a keynote at TechCrunch Disrupt, stating that OpenAI is now the preferred partner for powering the Replit Agent.
The move marks a significant operational shift for the company, which previously relied heavily on Anthropic’s Claude models to run its platform. Masad explained that the decision was driven by a combination of cost, partnership dynamics, and performance metrics. The company uses internal benchmarks to assess how well different models handle the creation of applications from scratch.
Performance and Cost Drivers
During his presentation, Masad highlighted that Replit’s internal testing consistently showed superior results for OpenAI’s models in building vibe-coded applications. He noted that the company also publishes public benchmarks to track these metrics.
“I think ChatGPT is better now,” Masad said. He added that the company evaluates performance based on how effectively the models build applications from scratch. The shift to OpenAI was described as a strategic choice based on both economics and the strength of the new partnership.
Before striking a deal with OpenAI in August, Replit was a top-five customer for Anthropic. That earlier partnership helped drive significant revenue growth for the coding platform. However, the new default setting means most users will now interact with OpenAI’s technology rather than Anthropic’s, likely reducing Replit’s spending with the latter.
Model Agnosticism Remains
Despite the change in default settings, Replit maintains that it remains model agnostic. Customers can still access Anthropic models on the platform, but they must actively select them rather than using the pre-set option. This distinction is important because most users tend to stick with the default model provided by the interface.
After his Disrupt talk, Masad clarified the situation to reporters. He stated that while OpenAI is the preferred partner for the Replit Agent, users retain the freedom to choose Anthropic’s Claude or other supported models. This approach allows the company to offer flexibility while optimizing the default experience for the majority of its user base.
The announcement has drawn mixed reactions from industry observers. Insiders familiar with the previous Anthropic partnership suggested the relationship was still active. Evidence on Replit’s website and social media channels showed promotions for Anthropic products as recently as late September. Blog posts from that time cited Anthropic’s Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent’s production environment. A product employee also posted about using Anthropic’s flagship model Opus 5.5.
Amjad Masad also shared insights into Replit’s financial trajectory during the Disrupt event. He noted that enterprise contracts now account for a larger portion of the company’s revenue. Last year, enterprise business made up about 2% of revenue. This year, that figure has risen to approximately 15%, with more than 50% of total revenue coming from B2B sources.
Looking further ahead, Masad expects enterprise and government customers to contribute more than half of Replit’s revenue by 2027. He also discussed the importance of AI safety, comparing the need for secure environments for AI agents to the requirements of human workers. He emphasized that agents need a habitat with good orchestration, error recovery, and sandboxing to operate safely.
Source: TechCrunch
