Replit CEO Amjad Masad Says AI Coding Platform Has Ended Partnership With Anthropic

Replit Chief Executive Officer Amjad Masad announced on Thursday during an onstage panel at the TechCrunch Disrupt conference in San Francisco that the cloud software development platform is no longer working with San Francisco AI safety and research company Anthropic. Speaking before technology investors and software engineers at Moscone West on Howard Street, Masad pointed to a combination of model inference cost, commercial partnership terms, and coding performance metrics as the primary drivers behind the strategic shift. The executive's remarks highlighted intensifying competition among frontier foundation model providers in the Bay Area.

During the conference session, Masad stated that OpenAI's ChatGPT architecture currently performs better on Replit's internal software engineering benchmarks, which the company uses to evaluate automated code generation and debugging accuracy. Replit previously formalized a strategic enterprise partnership with OpenAI in August to expand model integration across its developer workspace. That shift coincided with rapid commercial growth for the Bay Area coding platform, where enterprise business-to-business contracts now account for more than 50 percent of total company revenue, up from 2 percent of enterprise share last year to 15 percent this year.

Despite the chief executive's public declaration on the Disrupt stage, an unnamed industry insider familiar with the commercial arrangement disputed the characterization and indicated that the partnership between Replit and Anthropic remained in place. Observers also pointed out that an official engineering blog post published by Replit in late September explicitly cited Anthropic's Claude Fable 5 and Claude Fable 5.1 models as active components running inside Replit Agent's production environment.

Enterprise software analysts and venture capital board advisors across San Francisco, Palo Alto, and Oakland noted that AI application platforms increasingly rely on multi-model routing committees to balance token inference budgets against specialized coding capabilities. As corporate engineering departments adopt automated coding agents, platform vendors face growing pressure to negotiate favorable volume pricing and service-level agreements with frontier laboratory suppliers.

Software developers, enterprise technology directors, and industry researchers can review Replit's current production model documentation and enterprise security architecture online at replit.com or contact Moscone West conference organizers in San Francisco at (415) 974-4000. Further clarification regarding which foundation models power specific tiers of Replit Agent is expected in upcoming developer release notes.