Replit Switches Default Model to OpenAI, Drops Anthropic

Executive Overview

Replit CEO Amjad Masad announced at TechCrunch Disrupt on Thursday that OpenAI's ChatGPT is now the default model for Replit Agent, replacing Anthropic. The decision is based on cost, performance, and partnership. The change is effective immediately on the cloud platform, with Claude still available manually. Enterprise contracts have risen from 2% to 15% of revenue, and over 50% of revenue now comes from B2B, with expectations to exceed half from enterprises and government by 2027.

Deep-Dive Features & Architecture

The announcement came from Amjad Masad during a session at TechCrunch Disrupt, where he explained that Replit was until recently among Anthropic's top five customers, and the Claude partnership helped the company grow revenue significantly. However, the shift to OpenAI as the preferred partner for Replit Agent is based on three criteria: performance according to internal and public benchmarks, economics (cost), and partnership strength. Masad noted that OpenAI models named 'Sol' and 'Astra' consistently outperform in building 'vibe-coded' applications from scratch.

Architecturally, Replit maintains a 'model agnostic' approach, allowing users to manually select Claude or other supported models. But because most users stick with the default, this shift may significantly reduce Replit's spending with Anthropic. On the business side, Masad revealed that enterprise contracts rose from 2% of revenue last year to 15% this year, and over 50% of revenue now comes from B2B, with expectations that more than half of revenue will come from enterprises and government agencies by 2027.

Masad also stressed the need for a 'habitat' environment for intelligent agents including good orchestration, error recovery, security and sandboxing, networking, and identity and authorization, considering these elements essential for safe AI deployment.

Benchmark & Competitive Performance

The report does not include specific numerical benchmark figures, but indicates that Replit uses internal and public benchmarks to evaluate performance, and their results show OpenAI models (Sol and Astra) outperforming competitors in from-scratch building tasks. Meanwhile, sources familiar with the Anthropic partnership rejected Masad's statements, confirming the partnership continues, and noting that Replit was promoting Claude Fable 5, 5.1, and Opus 5.5 models just weeks ago. This discrepancy means the competitive comparison between OpenAI and Anthropic in the Replit context remains disputed among parties, and no independent numbers are available to settle superiority.

Industry Impact & Enterprise Adoption

For developers in the Arab world, Replit's shift to OpenAI as default means emerging 'vibe-coding' projects will likely rely on OpenAI models, which may change cost and efficiency calculations for processing Arabic text compared to Claude. Since the platform allows manual model selection, teams can still choose Claude for Arabic-specific tasks. The enterprise revenue growth from 2% to 15% and B2B exceeding 50% signals Replit's strengthening position in the enterprise market, with government and enterprise expected to drive over half of revenue by 2027. This shift also highlights the intensifying competition between AI model providers for platform partnerships, with cost and performance as key battlegrounds.

Conclusion

Replit's decision to make OpenAI the default model for Replit Agent, while keeping Anthropic available manually, reflects a strategic balance of cost, performance, and partnership. The enterprise revenue growth and B2B dominance underscore Replit's evolving business model. As the AI coding platform landscape evolves, the OpenAI-Anthropic rivalry will continue to shape developer tool choices, with users benefiting from model-agnostic flexibility.

Media Source: TechCrunch AI | Fact Verification & Analysis: AI Tools Oasis