Rednote, China’s top social app, has released its first open-source large language model. Dots.llm1 debuted on Hugging Face this week, marking a shift in how Chinese tech firms share AI. The move lets Rednote showcase its AI prowess and build a global developer community. It also counters U.S. export curbs on advanced semiconductors. With 300 million monthly users and a valuation near US$ 26 billion, Rednote aims to boost its international profile through transparent AI tools. The decision could reshape competition with U.S. giants that mostly keep their best models proprietary.
Inside Rednote Humane Lab: A Lighter, Smarter AI Model
Dots.llm1 is a mixture-of-experts model that activates 14 billion parameters out of 142 billion per query. The design matches top AI performance while cutting compute costs. Rednote’s Humane Intelligence Lab developed it using 11.2 trillion non-synthetic tokens, the company said. “We built dots.llm1 for cost-efficient performance and transparency,” said a lab researcher.
The team chose a mixture-of-experts structure to lower training and inference expenses. By sharing model checkpoints every trillion tokens, Rednote invites outside researchers to study its learning curve. The lab also released a detailed technical paper on Friday, outlining model design, training methods, and benchmark results on coding tasks.
Outpacing Rivals or Catching Up? Early Signs and Speed Bumps
In coding benchmarks, dots.llm1 matches Alibaba’s Qwen 2.5 series but trails DeepSeek-V3, the paper notes. Early adopters on Hugging Face have downloaded the model over 20 thousand times. Developers praised its performance on Chinese-language tasks and low latency. Yet some users report occasional errors in long replies.
Rednote has not used synthetic data, focusing on high-quality native text. That choice may limit training diversity but boost language precision. The company this year opened an office in Hong Kong’s Causeway Bay to expand internationally. It is also trialing Diandian, an AI search assistant on its platform. Future plans include further fine-tuning and tuning for vertical tasks such as e-commerce recommendations.
Rednote Openness Pushes China’s AI Strategy Forward
By open-sourcing dots.llm1, Rednote joins Alibaba, Meta, and DeepSeek in the global AI open-source wave. This strategy may strengthen China’s influence in AI research and speed developer adoption. As export restrictions challenge hardware access, software openness offers a new path to leadership.
Rednote’s transparent approach could spur collaboration across borders and invite scrutiny of its alignment with human values. If other firms follow suit, developers worldwide may gain access to more diverse models. Rednote plans to refine dots.llm1 further and release new checkpoints as training progresses. Its next step will be integrating community feedback into future releases, shaping AI innovation for a broader audience.
About The Author
