Replit, the AI-powered coding platform used by startup founders and developers, has stopped its collaboration with Anthropic, CEO and co-founder Amjad Masad announced Thursday during an appearance at TechCrunch Disrupt. The split comes months after Replit revealed a deal with OpenAI in August, though the company never publicly disclosed it was ending its relationship with the AI lab behind Claude. Masad pointed to multiple factors driving the decision, including expenses and how well the models performed.
Under the former arrangement with Anthropic, Replit relied on Claude models to power its cloud-based software development platform, which lets users build and launch applications straight from a web browser. That collaboration reportedly helped the company boost its revenue considerably. However, people with knowledge of the arrangement between the two companies disputed Masad's characterization, with one anonymous source saying the partnership remained active. Evidence on Replit's digital properties suggests the company was still tapping Anthropic's technology as recently as late September, when a blog entry referenced Claude Fable 5 and Claude Fable 5.1 as models operating in Replit Agent's live setting. Around that same period, a Replit product team member posted on X about deploying Anthropic's top-tier Opus 5.5 model and highlighting what it could do.
"I think ChatGPT is better now," Masad said, explaining that Replit applies an internal benchmark to evaluate metrics like performance. He added that the company releases a public benchmark as well, and its Sol and Astra products consistently excel at creating vibe-coded applications from the ground up. The choice came down to expense, partnership dynamics, and performance, Masad noted, saying OpenAI felt like the stronger partner for the moment. During his talk at Disrupt, Masad also shared that enterprise customers now represent a much bigger portion of revenue. Last year, enterprise made up roughly 2% of sales, he said, while this year it accounts for about 15%, with business-to-business contracts comprising more than half of total revenue.
The shift reflects Replit's evolving strategy as it weighs which AI providers deliver the best combination of capability and value. Masad also addressed the importance of safety infrastructure for AI systems, comparing it to the environmental requirements humans need to work effectively. Agents require what he called a habitat with strong orchestration, error recovery, safety and security features, sandboxing, networking, identity controls, and authorization, he explained. Once those elements are sorted out, companies can roll out AI in a secure manner, he said. The confusion over whether the Anthropic partnership is truly over or still ongoing remains unresolved, with TechCrunch noting it reached out to Replit for clarification. For platforms like Replit that depend on third-party AI models, the stakes involve not just raw performance but also the long-term reliability and economics of vendor relationships. Businesses watching this space will need to balance technical benchmarks against the risk of switching costs and partnership stability.