Replit makes OpenAI its default over Anthropic, CEO says

Replit, the AI coding platform beloved by startup founders and developers, is switching its main model provider. Co-founder and CEO Amjad Masad said on stage at TechCrunch Disrupt on Thursday that the company is no longer working with Anthropic, and that OpenAI is now its default.

Masad cited several reasons: cost, the nature of the partnership, and performance. "I think ChatGPT is better now," he said, pointing to an internal benchmark. On the company's public benchmark, he added, the Sol and Astra models consistently do better at building vibe-coded applications from scratch.

Anthropic hasn't been cut out entirely. After the talk, Masad clarified that customers can still pick Anthropic's models, but ChatGPT is now the default. Since most users stick with the default, that could meaningfully shrink Replit's spend with Anthropic.

The platform previously ran on Claude, and that partnership helped grow revenue significantly. Masad noted Replit had, until recently, been a top-five customer of Anthropic's. The OpenAI deal was announced back in August, but there had been no sign the Anthropic relationship was ending.

Masad's on-stage remarks were disputed by people familiar with the partnership. One, speaking anonymously, said the Anthropic deal was still in place. Replit's website, blog, and social posts were promoting Anthropic products just weeks ago: a late-September post lists Claude Fable 5 and Fable 5.1 as models running in Replit Agent's production, and an employee praised the flagship Opus 5.5 on X.

Masad also shared growth figures: enterprise contracts rose from 2% of revenue a year ago to 15% now, with more than half of revenue coming from B2B. In 2027, he expects over half of income to come from enterprise and government customers.

He touched on AI safety too. Like people, he argued, agents need a comfortable "habitat": good orchestration, error recovery, sandboxing, networking, and identity and authorization. "And once you figure all of that out, then you can deploy AI safely," he concluded.

Source: TechCrunch