Key takeaways
- Replit CEO announced OpenAI's ChatGPT is now the default model in Replit Agent, though Anthropic's Claude remains available as an option.
- The shift was driven by performance benchmarks, cost considerations, and partnership strength, with Replit previously being Anthropic's top-five customer.
- Insiders disputed the partnership claim, and Replit's blog and social posts from late September still promoted Anthropic models, creating confusion about the true status.
- Enterprise revenue grew from 2% of Replit's revenue last year to 15% this year, with the company expecting over 50% from enterprise and government by 2027.
Replit, the cloud-based coding platform, has deprioritized Anthropic as its primary AI model provider in favor of OpenAI, CEO Amjad Masad announced Thursday during a stage appearance at TechCrunch Disrupt. The shift makes OpenAI’s ChatGPT the default model powering Replit Agent, the platform’s AI-powered coding assistant, marking a significant realignment in how one of the developer community’s most popular tools sources its artificial intelligence.
The Announcement and Practical Shift
Replit had announced its partnership with OpenAI in August, but the decision to move away from Anthropic was not previously disclosed publicly. Masad revealed the change while speaking at the conference, citing cost, performance metrics, and partnership strength as the drivers behind the decision.
For years prior, Replit relied on Anthropic’s Claude models as its core AI backbone. The partnership reportedly delivered substantial revenue growth for Replit, and Masad acknowledged during his stage appearance that Replit had been among Anthropic’s top-five customers. The default shift from Claude to ChatGPT carries practical consequences: although customers retain the ability to manually select Anthropic’s Claude models or other supported alternatives, the vast majority of users stay with preset defaults rather than switching options themselves. A platform-wide default change typically translates to a significant reduction in spending with the previous provider.
What Remains Available
The announcement created immediate ambiguity about Anthropic’s status. Following the conference, Masad clarified to reporters that “Replit remains model agnostic. Our users can choose Anthropic’s Claude or other supported models.” This phrasing suggests Anthropic’s Claude remains an option within Replit’s ecosystem, but with diminished visibility and usage compared to the new default architecture.
Performance and Economics Drive the Decision
Masad stated that Replit’s internal benchmarking showed ChatGPT outperforming Anthropic’s models, particularly when building applications from scratch. “ChatGPT is better now,” he said on stage, noting that Replit also publishes a public benchmark that consistently shows OpenAI’s Sol and Astra models delivering stronger results on this metric. Alongside performance claims, he cited economic factors and the strength of the partnership relationship as contributing elements to the strategic shift toward OpenAI.
Partnership Strength as Factor
The partnership dimension appears to have played a role alongside pure performance metrics. Companies often factor commercial terms, product roadmap alignment, and relationship quality into infrastructure decisions. Masad emphasized all three considerations shaped Replit’s conclusion that OpenAI represented “a better partner for now,” language suggesting the decision was not solely driven by benchmark numbers.
Contradiction With Insiders and Recent Marketing
The announcement faced immediate pushback from people with direct knowledge of Replit’s arrangements with Anthropic. Insiders familiar with the partnership told reporters the relationship remained active, contradicting Masad’s public characterization of the partnership ending. The discrepancy raises questions about whether the partnership truly concluded or whether its scope narrowed without formally terminating.
Evidence on Replit’s own platforms muddied the picture further. A blog post published in late September cited Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent’s production environment. Around the same time, a Replit product employee posted on X platform promoting Anthropic’s flagship Opus 5.5 model and highlighting its capabilities. Both posts came after the OpenAI partnership announcement but weeks before Masad’s Disrupt comments, suggesting Anthropic remained integrated into Replit’s product narrative until recently.
Enterprise Revenue Acceleration Reshapes Strategy
Beyond the partnership announcement, Masad shared insight into Replit’s revenue composition, revealing a significant strategic pivot toward enterprise and government customers.
Growth in B2B Revenue
Enterprise contracts accounted for just 2% of Replit’s revenue one year ago. This year, that proportion has climbed to 15%, with more than 50% of total revenue now flowing from business-to-business relationships. Masad signaled this trajectory will accelerate, expecting enterprise and government customers to provide more than half of Replit’s revenue by 2027.
Implications for Model Selection
The enterprise focus provides context for the OpenAI preference. Enterprise buyers often operate within existing technology ecosystems and procurement relationships, many already built around OpenAI products and APIs. Optimizing Replit’s default for ChatGPT aligns the platform with common enterprise standardization patterns, even though individual developers retain the flexibility to choose alternative models.
Safety Infrastructure as Competitive Frame
Masad outlined his perspective on safely deploying AI agents at scale, framing it as a core platform capability. He drew an analogy between human work environments and the infrastructure requirements for autonomous AI systems, arguing agents need orchestration, error recovery, safety protections, security measures, sandboxing, networking, and identity and authorization systems to function reliably.
“Once you figure all of that out, then you can deploy AI safely,” Masad said. This framing positions Replit not merely as a consumer of AI model capabilities but as the steward of the operational and safety infrastructure required for agent deployment, an increasingly central concern as AI capabilities expand.
Source: TechCrunch
Frequently Asked Questions
Can Replit users still access Anthropic's Claude models?
Yes. Although OpenAI's ChatGPT is now the default model in Replit Agent, users can manually select Anthropic's Claude or other supported models. However, most users stick with the default option rather than switching.
Why did Replit switch to OpenAI as the default?
Replit CEO Amjad Masad cited three reasons: Replit's internal benchmarks showed ChatGPT performing better for building applications from scratch, cost considerations, and the strength of the partnership relationship with OpenAI.
What did insiders say about the Anthropic partnership?
Insiders familiar with the partnership told reporters it remained active, contradicting Masad's public statements that Replit no longer works with Anthropic. Additionally, Replit's blog posts and employee social media posts from late September still promoted Anthropic models like Claude Fable 5.1 and Opus 5.5.
