Chinese AI Labs Release New Open-Weight Models, Focus on Agent Runtimes

Several Chinese AI labs have released new open-weight models, including Z.ai's GLM-5.3, which is focused on coding and cybersecurity and was developed through scaled post-training. Alibaba's Qwen3.8-27B is a multimodal model with a large context window, designed for practical applications like coding and office workflows. DeepSeek V4-Pro and RedNote's dots3-note also continue this trend, with dots3-note introducing a new reinforcement learning method for long-horizon tasks. Discussions are also highlighting the importance of agent runtimes and harnesses, with DeepSeek Harness being treated as infrastructure that allows for modularity and dynamic component swapping. AI

IMPACT Accelerates the open-weight model ecosystem and highlights advancements in agent runtime infrastructure.

RANK_REASON Multiple frontier-lab model releases with system cards and benchmark results. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on Smol AINews →

• Alibaba
• Cursor
• DeepSeek
• DeepSeek Harness
• DeepSeek V4-Pro
• dots3-note
• GLM-5.3
• MiniMax
• Moonshot
• Qwen3.8-27B
• X.ai
• Z.ai

• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →