Replit Builds Its Own Coding Model, Breaking Reliance on OpenAI

Replit closed a $97.4 million funding round and unveiled a proprietary large language model for code that outperforms OpenAI Codex on selected benchmarks. The move signals vertical integration: Replit now controls its own inference layer rather than reselling third-party capacity. That shift gives the platform tighter command over latency, fine-tuning velocity, and margins—material advantages at Replit's scale across millions of concurrent developer sessions.