The performance gap between U.S. closed-source frontier AI models and Chinese open-weight models has narrowed to approximately 4.4 months, Mozilla’s latest State of Open Source AI report said. Open-weight models represented by Moonshot AI’s Kimi K3 trail Anthropic Fable 5 by only three points on the Artificial Analysis Intelligence Index composite score, while costing about 30% as much. METR estimates show the strongest closed-source model can reliably complete human expert tasks lasting about 12 hours, compared with about 7 hours for the strongest open-source model. Mozilla expects open-source models to catch up to the current closed-source level in about four months and said most organizations should adopt open-source models by default, paying a premium for closed-source models only in specialized scenarios requiring high-intensity retrieval or long context.