[Xiaohongshu AI Team Open-Sources dots3-note, Terminal-Bench 2.1 Score: 75.1] The Xiaohongshu AI Lab, dots.studio, has announced the open-sourcing of dots3-note preview. This model utilizes a 280 billion parameter MoE architecture with 16 billion active parameters, offering a 512,000-token context window and supporting multimodal understanding of text, vision, and audio. It introduces the TEMPO reinforcement learning method for long-term agent training. The model weights are now available on Hugging Face, and the API has been integrated with OpenRouter. Charts shared by SemiAnalysis show that dots3-note scored 75.1 on Terminal-Bench 2.1, outperforming the best-performing U.S. open-weight model in the chart by 4.9 points.