← All models

Nemotron-H 4B Instruct 128K

NVIDIA · 2025-10-23 · 4.5B parameters

Nemotron-H 4B Instruct 128K is NVIDIA's open-weight 4B hybrid Mamba-Transformer chat model aligned from Nemotron-H 4B Base 8K for single- and multi-turn chat, instruction following and tool calling; the BF16 checkpoint contains exactly 4,489,223,040 parameters. It was trained with staged supervised fine-tuning for math, code, science, chat and tool calling followed by RPO preference tuning, extending the context to 131,072 tokens. The card dates its public Hugging Face release to October 23, 2025, and the weights are released under the NVIDIA Internal Scientific Research and Development Model License.

Benchmark scores