Exclusive: Paying for frontier AI models buys 4-month head start at 5x the cost
Summary
Mozilla's State of Open Source AI shows the performance gap between frontier (closed) models and open-weight models has narrowed to about 4.4 months, driving many organizations to favor open models for routine work while reserving closed models for specialized tasks. Moonshot AI's Kimi K3 is shown to be only three points behind Anthropic's Fable 5 while costing roughly 30% as much. The article discusses when the premium for closed models is worth it, market dynamics around open-source AI, and calls for a broader open ecosystem.