China’s leading open-weight artificial intelligence models are now just 4.4 months behind the most advanced closed AI models from the US, according to Mozilla’s latest State of Open Source AI report.
The finding highlights how quickly Chinese AI companies have narrowed the gap with US technology firms while offering models that can be downloaded and run by developers and businesses.
Performance Gap Has Shrunk Dramatically
Mozilla’s analysis uses data from METR, which measures AI capability by looking at the length of tasks a model can reliably complete.
The latest assessment puts the gap between leading open-weight and closed frontier models at around 4.4 months. An independent estimate from Epoch AI puts the gap at roughly four months.
The gap has narrowed considerably compared with earlier years, when open models generally lagged proprietary systems by much longer periods.
Chinese Companies Dominate Open Models
Chinese AI companies are responsible for much of the progress in open-weight models.
Companies including Moonshot AI, Alibaba, DeepSeek and Z.ai have released models that are increasingly competitive with leading proprietary systems from US companies.
Moonshot AI’s Kimi K3, Alibaba’s Qwen 3.8 Max and DeepSeek’s V4-Pro-0813 are among the leading open models identified in the latest analysis.
Mozilla’s report also found that eight of the top 10 models by token volume on OpenRouter in August were open-weight models, with most of those developed by Chinese companies.
Open Models Cost Much Less
The biggest advantage of open-weight AI may not be performance alone but cost.
Mozilla’s analysis found that some open models can deliver comparable performance at a fraction of the price of closed alternatives. In one comparison, Z.ai’s GLM-5.2 performed within one point of Anthropic’s Claude Opus 4.7 on a particular evaluation while costing around five times less per completed task.
This makes open models increasingly attractive for routine workloads such as coding, classification, summarisation and other repetitive tasks.
Closed Models Still Have an Advantage
The narrowing gap does not mean open models have completely caught up.
Closed frontier models continue to have advantages on certain demanding workloads, particularly complex agentic tasks, long-context applications and highly specialised professional work.
They also typically come with infrastructure, support, compliance features and ready-to-use tools that businesses may find difficult to reproduce when operating open models themselves.
AI Industry Faces a New Competitive Landscape
The rapid progress of open models is changing the economics of artificial intelligence. Businesses can increasingly choose between expensive proprietary systems and cheaper open models that can be deployed on their own infrastructure.
Mozilla has argued that the growing concentration of the open-model ecosystem in China also raises questions about the need for more open-model development in the US, Europe and other regions.
Summary
China’s leading open-weight AI models are now estimated to be just 4.4 months behind US frontier models, according to Mozilla. Chinese companies dominate the open-model landscape, with Kimi, Qwen, DeepSeek and GLM among the leading systems. Their lower costs are encouraging businesses to use open models for routine AI workloads, although closed models retain advantages on some complex tasks.
