Village's LLM Atlas analysis shows closed AI models now control 97% of the price-performance frontier across 201 logarithmic budget points, reducing open-weight models' share to a historic low of 3%, concentrated solely at the cheapest tier via Google's Gemma 4 E4B and Xiaomi's MiMo-V2.6-Flash. This marks a sharp decline from June 2026, when open models held 41% of budgets, led by DeepSeek V4 Flash (nearly one-third) and Zhipu's GLM-5.2. The reversal followed OpenAI's September GPT-6 Luna and October GPT-6.1 Terra releases, which, alongside Anthropic's Claude Haiku 5.5, recaptured the mid-range. Historically, open models peaked episodically — DeepSeek-V3.2-Exp captured nearly half of budgets at its September 2025 launch — but shares halved by the next quarter as closed labs undercut them. The study attributes closed labs' sustained advantage to three factors: in-house distillation of flagship models (e.g., Haiku 5.5 released two weeks after Opus 5.5), comprehensive model families covering 11 of 21 frontier points (GPT-6/6.1), and lower inference costs from massive scale and custom hardware (Google TPUs). Methodology uses capability scores from 10+ benchmark families and API-derived costs relative to GPT-5 high-effort; restricting budgets to 3–100% of GPT-5 cost preserves the 3% open share, while extending to 0.3% raises it to 16% due to tiny open models.