k.Frontier
FrontierMarkTechPost
KeyRank 78NVIDIA AI Releases Star Elastic: One Checkpoint that Contains 30B, 23B, and 12B Reasoning Models with Zero-Shot Slicing
NVIDIA demonstrates a fundamental shift in model scaling efficiency: instead of training separate variants, one post-training method now produces multiple nested models from a single run. This directly impacts inference cost and deployment flexibility for enterprises choosing between model sizes.
Why it ranks · · Star Elastic enables 30B, 23B, and 12B models in single checkpoint via zero-shot slicing · 2026-05-09
Read full story