Jonathan Cohen, Nvidia's vice president of applied research, named the three at NVIDIA AI Day Seoul 2026 in a session after Deepu Talla's physical AI talk.
They are the last teams in a Ministry of Science and ICT project launched to cut Korea's reliance on foreign AI models, with two winners due early next year.
Nemotron is Nvidia's family of open-weight models, which the company releases along with training datasets, recipes and software.
Nemotron 3.5 Super VL, a 120-billion-parameter model released Oct. 15, is about as capable as the 550-billion-parameter Nemotron 3 Ultra but runs about five times faster, according to Cohen.
Nvidia launched the Nemotron Coalition in March to pool data and evaluations from AI labs including Mistral AI and Perplexity. Naver, whose cloud unit was dropped from the government project in January, joined in June as the coalition's first Korean member.
AI agents, programs that break a job into steps and hand each step to a model or tool, are where those models go to work. Cohen played down the mystique.
"It's just software. It's just the latest iteration in many, many decades of continual evolution," he said.
Nvidia's routing tools send easy tasks to cheaper models, NIM containers package models for deployment and its open-source Dynamo software schedules inference across a data center.
Agents barred from the internet but free to talk to one another have teamed up to slip those limits, a risk Nvidia's OpenShell sandbox targets by keeping credentials out of their reach.
Cohen cast Nemotron as an open foundation for the whole industry to build on.
"We plan to invest in the Nemotron family forever," he said.
AJP Takeaways
- Nvidia Vice President Jonathan Cohen said on Nov. 9, 2026, at NVIDIA AI Day Seoul 2026 that SK Telecom, LG AI Research and Upstage are using datasets from Nvidia's open Nemotron models.
- The three companies are the remaining teams in South Korea's Ministry of Science and ICT sovereign AI foundation model project, which advanced them after an Aug. 18, 2026, evaluation and will select two winners in early 2027.
- Nvidia released Nemotron 3.5 Super VL, a 120-billion-parameter open model, on Oct. 15, 2026.
Copyright ⓒ Aju Press All rights reserved.