Bloomberg has reported that Cursor, the AI coding startup behind Composer 2, is developing a proprietary frontier model designed to compete directly with Anthropic and OpenAI on coding tasks. The move signals a strategic shift from being a pure application layer on top of third-party models to becoming a vertically integrated AI coding company.
The decision makes commercial sense given Cursor's scale. At $2 billion ARR, the company is paying substantial inference costs to model providers. Training a competitive in-house model would reduce those costs while giving Cursor full control over the model's behaviour, fine-tuning, and optimisation for coding-specific tasks.
For the broader AI coding ecosystem, this is a significant development. It means the line between AI coding tools and AI model providers is blurring. Cursor would join a small group of companies — alongside Anthropic with Claude Code and OpenAI with Codex — that control both the model and the developer experience. The implications for pricing, performance, and competitive dynamics could reshape how developers choose their AI coding tools.
