Mozilla report: Chinese open-weight AI models trail US frontier by four months at a fraction of the cost
Mozilla's State of Open Source AI report, released September 15, finds that the best Chinese open-weight models lag leading closed models by about four months on task-horizon benchmarks. The top open model scored within three points of the closed leader on the Artificial Analysis Intelligence Index while costing 60% as much, and two points behind Claude Fable 5 at 30% of the price. The report is based on a survey of 1,400 developers and third-party benchmark data.
Mozilla’s report, based on a survey of 1,400 developers and third-party data, shows Chinese open-weight models like Z.ai’s GLM-5.2 scoring within a point of Claude Opus 4.7 on Terminal-Bench 2.1, at under one-fifth the cost per test. OpenRouter traffic data reveals eight of the top ten models by August token volume were open-weight, seven Chinese-built, yet closed providers captured 96% of model-layer revenue from May–September 2025. The report notes the four-month gap and 30% token price are measured API-to-API at list price; hardware-constrained deployments show larger drops—10 points for a single-server model and 23 for a single-GPU model versus the top closed model.
This narrowing gap could reshape enterprise AI procurement, as cost-sensitive developers may shift workloads to open-weight Chinese models without sacrificing much capability. However, the revenue disparity suggests trust and integration still favor closed providers, potentially slowing adoption. The four-month lag may compress further, pressuring US frontier labs on pricing and openness. Society could benefit from cheaper AI tools, but reliance on Chinese models raises questions about data governance and geopolitical dependencies, affecting regulators and businesses alike.