December 10, 2025 Chinese developers now hold most of the top positions on major community leaderboards that track the performance of open-source artificial intelligence models. A current snapshot from Hugging Face and shared on Reddit shows Chinese models taking the majority of the top 10 spots, led by systems from Alibaba’s Qwen family, Baidu’s ERNIE series, Tencent’s Hunyuan models, ByteDance research groups and several academic labs.
DeepSeek has added to this momentum with the release of two new models, including a reasoning system that early community testers say performs at a level comparable to ChatGPT 5.1 on a range of evaluation tasks. Publicly available model weights and documentation for the new releases are posted on DeepSeek’s GitHub repositories, including DeepSeek-R1 (reasoning model) and DeepSeek-V3 (general-purpose LLM).
Community leaderboards such as the Open LLM Leaderboard show the same trend: Chinese models now dominate upper-tier rankings across reasoning, coding and general-purpose benchmarks. These results are based on public evaluations and user-run tests rather than vendor claims.
China’s rapid progress is unfolding against a shifting hardware backdrop. Labs have been encouraged to design models that run efficiently on standard CPUs because of U.S. export controls on advanced GPUs. Yet reporting over the past year indicates that many of the highest-performing Chinese models are still trained on Nvidia accelerators acquired through distributors and grey-market channels. At the same time, domestic chip development is accelerating. Huawei’s Ascend line of AI processors is reported to deliver performance approaching Nvidia’s data-centre GPUs for certain workloads, and analysts say reliable mass production could further speed China’s model development cycle.
Researchers note that while U.S. companies continue to lead in proprietary frontier systems, China is moving quickly in the open-source segment that underpins academic research, startup development and low-cost enterprise adoption. Community testers say the growing capability of Chinese open models, including DeepSeek’s latest releases, suggests the gap between open-source and frontier models may narrow faster than expected.
Nvidia CEO Jensen Huang added to this last week, saying China’s AI capabilities are more advanced than many in the West realise and that underestimating that progress would be a mistake.
