NVIDIA Dynamo Recipes Bring Korea’s Leading Open Models to Production

NVIDIA has announced deployment recipes tuned for its Dynamo software and aimed at four prominent open models from South Korea. In a blog post dated November 8, 2026, and written by Amr Elmeleegy, the company names A.X K2 from SK Telecom, K-EXAONE 2.0 from LG AI Research, Solar Open 2 from Upstage and Motif-3 from Motif Technologies as the models covered.

Dynamo is NVIDIA's open source framework for serving inference across multiple nodes at data center scale. The company explains that once a model outgrows a single GPU, it has to be divided among several, and requests and communication between those chips must be carefully coordinated to keep performance up. Dynamo is designed to manage that routing and communication.

The new recipes are optimized for NVIDIA's B200 GPUs, and NVIDIA claims they deliver as much as twice the performance without additional tuning, removing what it calls "the guesswork" from deployment. The post offers no benchmark details or test conditions behind that figure.

The company frames the release as a way for developers to move Korean open models into production more quickly and turn them into widely available AI services. It appeared alongside other NVIDIA announcements the same day on Korea, covering physical AI, agentic AI and sovereign AI collaboration, which suggests the recipes are one part of a broader push by the company in the country's AI sector.