RedNote and Jeonnam-Gwangju are building AI in house
RedNote released an open-weight model with 280 billion parameters, while Jeonnam-Gwangju's civil servants built an assistant that processes festival data, bringing in-house AI development into routine public work.
Artificial Intelligence··Evening
RedNote opens its 280-billion-parameter model
RedNote's AI research unit Dots Studio released Dots3-Note Preview, an open-weight model with 280 billion parameters. The company says the model matched or exceeded systems from OpenAI, Anthropic, DeepSeek and Z.ai on specific tasks. That comparison rests on RedNote's own measurements, and the report includes no independent test. Dots Studio presents the model as a tool for solving everyday problems; releasing the weights also gives the company room to adapt its own system to products instead of relying only on an outside service.[1]
Consumer platforms move foundation models in house
The South China Morning Post places RedNote's release within a wider shift by Chinese consumer platforms toward building their own foundation models. Companies in e-commerce, gaming, social media and travel are no longer relying solely on systems bought from outside labs. Food-delivery and local-services platform Meituan is developing in-house models as well. Alongside AI-focused labs such as DeepSeek, Z.ai, Moonshot AI and MiniMax, platforms with large user bases and their own application data are therefore becoming model developers in their own right.[1]
City staff build their own assistant for festival data
The analysis assistant used in Jeonnam-Gwangju was not bought from an outside vendor; the city's data analysis team built it in 2026. Civil servants enter pre-verified base data through step-by-step prompts prepared for festivals and events. The generative model processes floating-population and card-spending figures and drafts the first report. The city's data analysis projects rose from 9 in 2023 to 27 in 2026, also covering airport routes, childcare, single-person households and financial vulnerability across 27 municipalities.[2]
