Replit the AI coding platform favored by startup founders and developers has stopped collaborating with Anthropic company CEO and co-founder Amjad Masad stated Thursday during an appearance at TechCrunch Disrupt.

Masad explained the choice stemmed from several factors including expenses. Replit revealed an alliance with OpenAI in August though nothing had signaled an end to its work with Anthropic.

ChatGPT now outperforms other options Masad noted while highlighting an internal benchmark used to track metrics such as speed. The firm also shares a public benchmark and Sol along with Astra show stronger results when creating vibe-coded apps from the ground up. Cost partnership details and performance all played roles leading the team to view OpenAI as the stronger fit at present.

Replit offers a cloud-based development platform that lets users build and launch apps straight from a browser. Its earlier work with Anthropic involved running Claude models on the platform and that arrangement reportedly boosted revenue growth.

Insiders close to the Anthropic and Replit partnership pushed back on Masad's statements. One person who requested anonymity said the arrangement continues without interruption.

Company website posts blog entries and social media activity show Replit relying on Anthropic tools as recently as several weeks ago. A late September blog entry references Claude Fable 5 and Claude Fable 5.1 running in Replit Agent production settings. Around that period a Replit product team member posted on X about testing Anthropic flagship model Opus 5.5 and highlighting its strengths.

TechCrunch contacted Replit again for further details and will provide updates upon receiving a response.

Masad discussed company growth during the Disrupt session noting that enterprise deals now represent a bigger share of overall revenue.

ARR figures remain private for now yet enterprise revenue stood at roughly 2 percent last year and has risen to about 15 percent this year with more than 50 percent of total revenue coming from B2B contracts Masad said.

Masad also addressed the need for strong AI safety practices.

Humans require suitable conditions to perform well such as a comfortable office good meals air conditioning and reliable computers he explained. Agents likewise need an environment that supports effective work. That setup includes solid orchestration reliable error recovery safety security sandboxing networking and proper identity controls. Once those elements are in place AI can be rolled out safely.

This article was updated to add further details on Replit use of Anthropic models.