Pattern: AI Coding Platforms Train Custom Models to Escape Foundation API Dependency

Cursor Blog·high signal

Cursor's Composer 2.5 (built on Kimi K2.5 with custom RL) matches Opus 4.7 at 1/10th token cost. This follows Cursor training Composer 2 and Codex building task-specific models on its own infrastructure. The pattern is clear: every major AI coding platform is investing in purpose-built post-trained models to reduce per-token costs and API dependency. For builders, the 'best model' for coding is increasingly not the general-purpose frontier model — it's a task-specific one.