Closed AI models now dominate the price-performance frontier

Closed AI models now hold nearly every point on the price-performance frontier, with open-weight models the best buy at just 3% of budgets, down from 41% in June, according to Village’s LLM Atlas. It marks the lowest open-model share measured since 2024, when open models held at least 6% of budgets at every quarter-end, with past peaks such as DeepSeek-V3.2-Exp’s near-half share in September 2025 fading by roughly half or more within a quarter. Village attributes the shift to closed labs distilling their own frontier models into cheaper tiers within weeks of flagship launches, releasing full model families—OpenAI’s GPT-6 and GPT-6.1 line holds 11 of the frontier’s 21 points—and serving inference at far greater scale, including Google on its own chips. Watch whether the next cheap open release, typically from DeepSeek, can hold a wide budget band for longer than a single quarter.