Pattern: AI Coding Model Commoditization — Open-Weight Fine-Tuning Matches Frontier at 1/10th Cost

Multiple (Cursor Blog, Google Developers Blog)·high signal

Cursor's Composer 2.5 proves the commoditization thesis: take an open-weight base (Kimi K2.5, 1T params), fine-tune on proprietary session data, and match Opus 4.7 on SWE-Bench at $0.50/M input tokens vs. $15. Google takes a different route with Gemini 3.5 Flash at 4x speed. The pattern: raw coding capability is no longer a moat — distribution, data flywheel, and agent orchestration are. Builders should expect pricing pressure across all AI coding tools through Q3 2026.