Mozilla Says Open AI Models Trail Frontier Systems by 4.4 Months as Costs Drop to 30%
Updated
Updated · Ars Technica · Sep 15
Mozilla Says Open AI Models Trail Frontier Systems by 4.4 Months as Costs Drop to 30%
1 articles · Updated · Ars Technica · Sep 15
Summary
Mozilla’s State of Open Source AI report says the lead of top US frontier models over the best Chinese open-weights models has narrowed to 4.4 months, making open models the default choice for most routine enterprise work.
Kimi K3 from Moonshot AI scored just 3 points behind Anthropic’s Fable 5 on the Artificial Analysis Intelligence Index while costing only 30% as much, illustrating why companies are shifting toward cheaper open alternatives.
Closed models still justify their premium in a narrower set of tasks—expert professional work, high-intensity retrieval and long-context use—rather than across an entire organization, Mozilla CTO Raffi Krikorian said.
Open-weights models let users download and run core model components themselves, but missing training data, pipelines and code—and the lack of built-in compliance, support and accountability—still keep many organizations paying for closed systems.