OpenAI Becomes Default Model
Replit has designated OpenAI’s ChatGPT as the default model for its platform, as announced by CEO and co-founder Amjad Masad during TechCrunch Disrupt. This shift is attributed to a combination of cost considerations, the strength of the partnership, and performance metrics assessed through internal benchmarks.
Reasons for the Change
Masad stated that Replit publishes a public benchmark and maintains internal performance metrics. He remarked, “I think ChatGPT is better now,” and noted that models named Sol and Astra outperform others in building vibe-coded applications from scratch. The decision was framed as one driven by performance, economics, and partnership.
Continued Access to Anthropic Models
Following the event, Masad clarified that while ChatGPT is now the default model, customers can still access Anthropic’s models on the Replit platform. This change is expected to reduce Replit’s spending with Anthropic, as most users tend to stick with the default model rather than selecting alternatives.
Partnership History and Discrepancies
Previously, Replit relied on Anthropic’s Claude models to operate parts of its platform, a relationship that Masad indicated contributed to the company’s revenue growth. He mentioned that Replit had been a top-five customer of Anthropic until recently. However, insiders familiar with the partnership disputed Masad’s comments, asserting that the collaboration with Anthropic remains intact.
Public Signals of Ongoing Collaboration
Evidence of continued collaboration with Anthropic includes a late-September blog post that cited Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent’s production environment. Additionally, a Replit product employee recently posted on X about utilizing Anthropic’s Opus 5.5 model.
Revenue Mix and Future Projections
Masad provided insights into Replit’s revenue mix, noting that enterprise revenue accounted for approximately 2% last year and is projected to rise to about 15% this year. He indicated that more than 50% of revenue now comes from B2B customers, with expectations that enterprise and government clients will represent over half of Replit’s revenue by 2027.
AI Safety Considerations
On the topic of AI safety, Masad drew a comparison between AI agents and human workers, emphasizing the need for an appropriate “habitat” for effective operation. He stated, “that habitat has good orchestration, good error recovery, good safety and security and sandboxing, good networking, good identity and authorization.”
In summary, while OpenAI’s ChatGPT is now the default model on Replit, users still retain access to Anthropic models, reflecting a nuanced partnership between the two companies.
Original source: TechCrunch AI