Open-Weight AI Models Narrow Gap with Frontier Models, Cutting Costs

A new Mozilla report released on September 15 shows that the performance difference between U. S. frontier AI models and the leading open‑weight models from China has shrunk to just 4. 4 months. The study finds that most routine tasks can be handled by open models, which cost only a fraction of their closed‑source counterparts. For example, Moonshot AI’s Kimi K3 scores only three points lower than Anthropic’s Fable 5 on the Artificial Analysis Intelligence Index, yet it is priced at roughly 30 % of the closed model’s cost. Mozilla’s chief technology officer, Raffi Krikorian, notes that the premium paid for closed models is justified only for a narrow set of workloads—expert professional work, high‑intensity retrieval, and long‑context processing. Consequently, organizations are advised to adopt open models as the default and reserve paid frontier models for those specific, high‑value use cases.