Mozilla Report: US-China AI Model Gap Narrows to 4.4 Months
Mozilla released the second edition of its 'State of Open Source AI' report, revealing that the performance gap between frontier U.S. proprietary models and leading Chinese open-weight models has narrowed to just 4.4 months. The report also highlights that open-weight models accounted for 8 of the top 10 models by token volume on OpenRouter in August 2026. The rapidly shrinking capability gap and significantly lower operational costs are encouraging enterprises to adopt open-weight models for everyday operations while reserving expensive proprietary models only for highly complex tasks. This shift is reshaping commercial AI adoption, giving developers powerful yet cost-effective alternatives to closed APIs. Proprietary models currently lead in reliable task duration, handling tasks up to ~12 hours compared to ~7 hours for open-weight models at a 50% success threshold. However, open models like Moonshot's Kimi K3 score within 3 points of Anthropic's Fable 5 while costing only around 30% as much.
## BACKGROUND
Open-weight AI models release their internal neural network weights, allowing developers to self-host, fine-tune, and run them locally or in private clouds rather than relying exclusively on vendor APIs. Router services like OpenRouter act as unified API hubs, letting developers dynamically send queries across hundreds of proprietary and open-weight LLMs based on cost, speed, and benchmark performance.