Replit, the cloud-based AI coding platform popular among developers and startup founders, has ended its partnership with Anthropic, CEO and co-founder Amjad Masad announced Thursday at TechCrunch Disrupt. The decision comes several months after Replit announced a new partnership with OpenAI in August, though the company had not previously disclosed the termination of its Anthropic collaboration. Masad attributed the move to multiple factors including cost considerations, partnership dynamics, and performance metrics.
Masad stated that ChatGPT now performs better according to Replit's internal benchmarks, which the company uses to evaluate metrics like application-building performance. He noted that Replit publishes public benchmarks showing that OpenAI's models consistently outperform competing solutions for constructing coded applications from scratch. The decision ultimately hinged on a combination of cost, partnership terms, and technical performance, with Masad concluding that OpenAI represented a superior partner for the company's current needs.
Under the Anthropic partnership, Replit had leveraged Claude models to power its platform, a relationship credited with driving significant revenue growth for the coding platform. The shift to OpenAI represents a meaningful pivot in the company's technical infrastructure and AI strategy.
The announcement generated some pushback from people familiar with the Anthropic-Replit relationship. An unnamed insider suggested the partnership remained active, contradicting Masad's public statements. Evidence emerged that Replit may have continued using Anthropic's products as recently as September, with a company blog post from that month specifically referencing Claude Fable 5 and Claude Fable 5.1 models running in Replit Agent's production environment. This apparent discrepancy raised questions about the exact timing and status of the transition between AI partners.
Beyond the Anthropic news, Masad highlighted significant shifts in Replit's revenue composition. Enterprise customers represented just 2% of revenue the previous year but have grown to account for 15% of current revenue. More broadly, B2B revenue now comprises more than 50% of total revenue, indicating a strategic expansion beyond individual developers toward organizational customers.
Masad also addressed the importance of safety infrastructure for deploying AI agents effectively. He framed proper AI safety measures as essential environmental conditions comparable to human workplace needs—drawing an analogy between office amenities like air conditioning and good food and the technical requirements for AI systems. He emphasized that effective AI deployment requires robust orchestration, error recovery, security sandboxing, networking capabilities, and identity and authorization controls. According to Masad, only after establishing this complete safety habitat can organizations responsibly deploy AI systems.
Gist is a free AI reader for your browser, iPhone, and Android. Get concise summaries and key takeaways from any article or podcast.
Get Gist — Free