Closed AI models dominate the price-performance frontier (village.dev)

A recent analysis from Village’s LLM Atlas reveals that closed AI models are now dominating the price-performance landscape, holding the top positions across nearly all budget tiers. While open-weight models had a brief resurgence, capturing about 41% of budgets in June, the release of OpenAI’s GPT-6 Luna and GPT-6.1 Terra in September and October has shifted the dynamics, pushing open models down to about 3% of budgets. This stark decline illustrates a trend where open models briefly excel in value but quickly lose ground to cheaper, more powerful releases from closed labs. The significance of this shift is profound for the AI and machine learning community, as it highlights the increasing efficiency and scale of closed labs like OpenAI and Anthropic. Three main factors contribute to their dominance: the ability to develop lower-cost, high-capacity models from their flagship offerings, the release of entire families of models that cover various price points, and the operational efficiencies gained from running inference on large scales using proprietary hardware. The methodology behind the analysis underscores the competition between closed and open models, indicating that while open labs have potential in lower price ranges, they currently struggle to maintain a foothold against the superior value propositions of their closed counterparts.