Xiaohongshu Model Outperforms US OpenAI by 4.9 Points

The Chinese AI laboratory Xiaohongshu, known as dots.studio, has unveiled its open-source model 'dots3-note preview'. This model features a MoE structure with 280 billion parameters and a context window of 512,000 tokens. The actual active parameters are reported to be 16 billion, and it supports understanding of not only text but also visual and audio inputs. Dots.studio has introduced the 'TEMPO' reinforcement learning method for long-term agent training. The model weights have been uploaded to Hugging Face, and the API is connected to OpenRouter. According to performance metrics, dots3-note achieved a score of 75.1 on the Terminal-Bench 2.1, surpassing the US OpenAI model by 4.9 points. SemiAnalysis is currently verifying the quality of everyday use and the potential for optimization in evaluations through additional testing. Terminal-Bench 2.1 is a benchmark for evaluating agents in terminal environments, revised from 28 out of 89 tasks. The MoE structure enhances computational efficiency by activating only certain expert models. This release is related to AI development infrastructure and provides developers and researchers the opportunity to compare and experiment with open-weight models through OpenRouter.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.