8 hours ago
- Business
- South China Morning Post
RedNote joins AI race with its own open-source model that it says bests Alibaba, DeepSeek
Chinese social media platform RedNote has open-sourced its first large language model (LLM), joining a growing list of Big Tech firms looking to stake a claim in the
artificial intelligence (AI) market, as the Shanghai-based company tries to leverage its growing international profile for new growth.
The company on Friday unveiled its model, a mixture-of-experts system that activates 14 billion parameters out of a total of 142 billion when responding to queries, a design that aims to match the performance of leading AI models while cutting training and inference costs. RedNote said the model was developed by its in-house Humane Intelligence Lab, or 'hi lab', which evolved from the company's previous AI research team.
RedNote, known as Xiaohongshu in Chinese, has 300 million monthly active users. Recent market transactions in the privately held company have seen
its valuation reach US$26 billion , surpassing its pandemic-era peak in 2021, Bloomberg reported last week. The company is expected to launch an initial public offering as soon as this year.
The company opened a new office in Hong Kong on June 7, its first outside mainland China, located at Times Square in Causeway Bay. RedNote has been looking to expand internationally this year after a boost in its popularity overseas when it looked like the US might ban
ByteDance 's
TikTok
China has seen a surge in the number of LLMs released in recent years, with a recent trend towards open-source models following the popularity of models from AI research firm
DeepSeek . Deep-pocketed tech giants such as
Alibaba Group Holding
Tencent Holdings and ByteDance have been leading the charge in training pricey foundational models. Alibaba owns the Post.
RedNote's Humane Intelligence Lab has this year been recruiting researchers with strong humanities backgrounds, emphasising the importance of humanlike expression and alignment with human values. The company claimed outperformed other leading open-source models in Chinese language understanding, including Alibaba's Qwen2.5-72B-Instruct and DeepSeek-V3.
