Jun 6, 02:15 PM

China’s Rednote Releases Dots.llm1 MoE AI Model With 14B Active Parameters, Trained on 11.2T Tokens; DMind Launches Web3 LLM on Alibaba Qwen3-32B

Chinese AI developer Rednote, associated with the social media platform Xiaohongshu, has released dots.llm1, an open-source large-scale Mixture of Experts (MoE) language model. The model activates 14 billion parameters out of a total 142 billion and was pretrained on 11.2 trillion high-quality, non-synthetic tokens. Dots.llm1 matches state-of-the-art performance comparable to Alibaba's Qwen2.5-72B and is distributed under an MIT license with checkpoints released every trillion tokens. This release highlights China's leading position in price-performance ratio for AI models. Separately, DMind has launched DMind-1, an open-source large language model built on Alibaba's Qwen3-32B, optimized for Web3 applications, offering superior benchmark performance at 10% of the cost. Its smaller variant, DMind-1-mini (14B parameters), ranks second, maintaining 95% of the power with smoother operation.