The Cost of Intelligence: Decoding Replit's Pivot to OpenAI

In the creator economy, the distance between a technical preference and a financial mandate is often razor-thin. For Replit, the cloud-based software development platform, the recent decision to sever ties with Anthropic is a clear signal that the company is prioritizing the bottom line and operational efficiency over legacy partnerships.

Speaking at TechCrunch Disrupt, Replit CEO and co-founder Amjad Masad confirmed that the company is no longer working with Anthropic. While Masad noted that the decision stemmed from a variety of factors, he explicitly cited cost as a primary driver. This shift comes on the heels of a partnership with OpenAI announced in August. According to Masad, the move was a calculation based on cost, partnership, and performance, stating that "ChatGPT is better now."

From a monetization perspective, this isn't just about which LLM is more capable; it is about the unit economics of AI-assisted coding. Masad revealed that Replit utilizes internal benchmarks to track performance, noting that OpenAI's Sol and Astra models consistently outperform others when building "vibe-coded applications from scratch."

***

**Opinion: The Burn Rate Pivot**

*In my view, this move is a textbook example of a platform optimizing its burn rate during a critical scaling phase. While the Anthropic partnership reportedly helped Replit grow its revenue significantly in the past, the current priority is clearly efficiency. As Replit shifts from a developer tool to an enterprise-grade platform, the cost per token becomes a primary lever for margin expansion.*

***

This financial tightening aligns with Replit's aggressive pivot toward the enterprise market. Masad shared that enterprise contracts, which accounted for only 2% of revenue last year, now make up 15% of the company's revenue. Furthermore, more than 50% of Replit's total revenue is now derived from B2B sources. As the company scales these high-value contracts, reducing the overhead of the underlying AI models is essential to maintaining sustainable growth.

However, the transition has not been without friction. TechCrunch reports that insiders familiar with the partnership dispute Masad's claims, suggesting the relationship with Anthropic is still in place. TechCrunch highlighted evidence from Replit's own blog and social media, noting that as recently as late September, a blog post cited Anthropic’s Claude Fable 5 and 5.1 in the Replit Agent’s production environment. Additionally, a Replit product employee recently promoted the capabilities of Anthropic’s Opus 5.5 on X.

Despite these contradictions, Masad's vision for the "self-driving company"—where AI agents handle the bulk of grunt work, such as triaging support tickets via GitHub and Linear—requires a lean, high-performance infrastructure. By moving toward OpenAI, Replit is betting that it can lower the cost of these agentic workflows while increasing the "taste" and judgment provided by human operators.