Rednote, one of China’s most popular social media platforms, has released its own large language AI model as open-source. This means anyone can use or build on it for free. Rednote is joining other Chinese tech companies that are also sharing their AI models with the public.
On Friday, Rednote introduced its new AI model called dots.llm1. It uses a “mixture-of-experts” system, which means it only activates 14 billion out of 142 billion total parameters when answering questions. This helps it work efficiently while keeping costs low.
The model was built by Rednote’s own research team, now called the Humane Intelligence Lab (or “hi lab”). This team grew out of Rednote’s earlier AI research group.
RedNote, known as Xiaohongshu in China, has 300 million active users each month. According to Bloomberg, its value has recently grown to $26 billion—higher than it was during the pandemic in 2021. The company may go public with an IPO as early as this year.
On June 7, RedNote opened a new office in Hong Kong—its first one outside mainland China. The office is in Times Square, Causeway Bay.
The company is planning to grow internationally, especially after gaining more attention overseas when there were talks about the U.S. possibly banning TikTok.
In recent years, China has seen a significant increase in the number of large language models (LLMs) being developed. Lately, more companies are choosing to make their models open-source, especially after the success of models from the AI firm DeepSeek.
Major tech companies like Alibaba, Tencent, and ByteDance are leading this trend. They are investing heavily in building robust and expensive AI models. (Note: Alibaba also owns the South China Morning Post.)
This year, RedNote’s Humane Intelligence Lab has been hiring researchers with strong backgrounds in humanities. They focus on making the AI express itself more like a human and follow human values.
RedNote says their dots.llm1 model works better at understanding the Chinese language than other popular open-source models, including Alibaba’s Qwen2.5-72B-Instruct and DeepSeek-V3.
RedNote is testing an AI research assistant called Diandian on its platform. They said Diandian uses a model created by their team, but they didn’t share the model’s name.
The company highlighted that they didn’t use any artificial or fake data to train the model. Instead, they trained it with 11.2 trillion pieces of high-quality real data.
The assistant, which includes a “deep research” function, is launched via a dialogue box within RedNote.
To be more open, RedNote is sharing parts of its AI model after every trillion tokens it learns. This lets outside researchers see how AI is improving over time.
Read more- AI-Driven Interiors Platform Roomstory.ai Raises INR 3 Crore in Funding
