China AI Native Industry Insights – 20260715 – Alibaba | StepFun | Tencent | more
Explore Qoder updates, StepFun’s Step Edge launch, and Tencent’s quantized model. Discover more in Today’s China AI Native Industry Insights.
1. Qoder JetBrains Plugin Major Update: Custom Agents, Computer Use, and Plan Mode Are Now Available
The Qoder JetBrains plugin has received a major upgrade, introducing Custom Agents, Computer Use, and Plan Mode to deliver a more powerful AI-assisted development experience. Developers can now create personalized agents for their own workflows, let AI generate structured execution plans for complex tasks, and automate desktop interactions through computer control. The update also brings improved conversation steering, better message handling, and a new cleanup tool for managing outdated Qoder files, making AI-powered coding workflows more flexible and efficient.
Read more: https://mp.weixin.qq.com/s/HgcAktKLkf5kg6-Bf3DHTQ
Video Credit: The original article
2. StepFun Launches Step Edge: A Full Suite of On-Device Models Bringing AI Agents Beyond the Cloud
StepFun has introduced Step Edge, a comprehensive family of on-device models designed to bring AI agents into real-world terminal environments such as smartphones and vehicles. The Step Edge lineup includes Step Edge Base Model, Step Edge Audio, Step Edge GUI, and Step Edge Gen, enabling faster local responses, enhanced privacy protection, and seamless cloud-device collaboration. With support for ultra-low-latency local tool execution, multimodal on-device processing, and optimization through the proprietary Step Inference NPU engine, Step Edge delivers a new generation of intelligent experiences where AI can perceive, reason, and act directly on devices.
Read more: https://mp.weixin.qq.com/s/StOzmXaUGSsjUXkAoW-HZg
Video Credit: NotebookLM
3. Tencent Hunyuan Hy3 Quantized: 295B-Parameter Flagship Model Now Runs on a Single GPU
Tencent Hunyuan has released highly optimized quantized versions of Hy3, enabling its 295B-parameter flagship model to run on significantly more accessible hardware. The team introduced 1-bit and 4-bit GGUF quantized models, along with a GPTQ Int4 version, reducing the original nearly 600GB BF16 model size to deployment-friendly formats. With llama.cpp integration, the Hy3 1-bit version can run on a single 96GB inference GPU, while the 4-bit version delivers near full-model performance with limited hardware resources. The release also adds MTP speculative decoding support, improving inference speed and making advanced capabilities such as agents, coding assistance, long-context understanding, and productivity workflows more practical for local deployment.
Read more: https://mp.weixin.qq.com/s/Kq30ftirASryPrUtjK2xSw
Video Credit: The original article
That’s all for today’s China AI Native Industry Insights. Join us at AI Native Foundation Membership Dashboard for the latest insights on AI Native, or follow our linkedin account at AI Native Foundation and our twitter account at AINativeF.