LLM EXPLORER 60,828 MODELS INDEXED

NVIDIA Nemotron 3 Super 120B A12B by unsloth

By unsloth · 294 downloads
HF Repo ↗ Copy URL

NVIDIA Nemotron 3 Super 120B A12B is an open-source language model by unsloth. Features: 120b LLM, VRAM: 214.5GB, Context: 256K, License: other, LLM Explorer Score: 0.37, ELO: 1361.

  Arxiv:2512.20848   Arxiv:2512.20856 Base model:finetune:nvidia/nvi... Base model:nvidia/nvidia-nemot...   Conversational   Custom code Dataset:nvidia/nemotron-post-t... Dataset:nvidia/nemotron-pre-tr...   De   En   Es   Fr   It   Ja   Latent-moe   Mtp   Nemotron-3   Nemotron h   Nvidia   Pytorch   Region:us   Safetensors   Sharded   Tensorflow   Zh
Model Card on HF 🤗: https://huggingface.co/unsloth/NVIDIA-Nemotron-3-Super-120B-A12B 

NVIDIA Nemotron 3 Super 120B A12B Benchmarks

LMSys ELO 1361 vs 1272 (so35) 7%
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

NVIDIA Nemotron 3 Super 120B A12B Parameters and Internals

LLM NameNVIDIA Nemotron 3 Super 120B A12B
Repository 🤗https://huggingface.co/unsloth/NVIDIA-Nemotron-3-Super-120B-A12B 
Base Model(s)  ...emotron 3 Super 120B A12B BF16   nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16
Model Size120b
Required VRAM214.5 GB
Updated2026-08-06
Maintainerunsloth
Model Typenemotron_h
Model Files  5.0 GB: 1-of-50   5.0 GB: 2-of-50   5.0 GB: 3-of-50   5.0 GB: 4-of-50   5.0 GB: 5-of-50   5.0 GB: 6-of-50   5.0 GB: 7-of-50   5.0 GB: 8-of-50   5.0 GB: 9-of-50   5.0 GB: 10-of-50   5.0 GB: 11-of-50   5.0 GB: 12-of-50   5.0 GB: 13-of-50   5.0 GB: 14-of-50   5.0 GB: 15-of-50   5.0 GB: 16-of-50   4.9 GB: 17-of-50   5.0 GB: 18-of-50   5.0 GB: 19-of-50   5.0 GB: 20-of-50   5.0 GB: 21-of-50   5.0 GB: 22-of-50   4.9 GB: 23-of-50   5.0 GB: 24-of-50   5.0 GB: 25-of-50   5.0 GB: 26-of-50   5.0 GB: 27-of-50   5.0 GB: 28-of-50   4.9 GB: 29-of-50   5.0 GB: 30-of-50   5.0 GB: 31-of-50   5.0 GB: 32-of-50   5.0 GB: 33-of-50   5.0 GB: 34-of-50   4.9 GB: 35-of-50   5.0 GB: 36-of-50   5.0 GB: 37-of-50   5.0 GB: 38-of-50   5.0 GB: 39-of-50   5.0 GB: 40-of-50   4.9 GB: 41-of-50   5.0 GB: 42-of-50   5.0 GB: 43-of-50
Supported Languagesen fr es it de ja zh
Model ArchitectureNemotronHForCausalLM
Licenseother
Context Length262144
Model Max Length262144
Transformers Version4.57.6
Vocabulary Size131072
Torch Data Typebfloat16

Best Alternatives to NVIDIA Nemotron 3 Super 120B A12B

Best Alternatives
Context / RAM
Downloads
Likes
...on 3 Super 120B A12B Base BF161024K / 209.5 GB1305332
...motron 3 Super 120B A12B NVFP4256K / 80.3 GB3041935408
...emotron 3 Super 120B A12B BF16256K / 194.6 GB864394414
...motron 3 Super 120B A12B NVFP4256K / 80.3 GB113227
...Nemotron 3 Super 120B A12B FP8256K / 128.4 GB8110
...Nemotron 3 Super 120B A12B FP8256K / 128.4 GB221272269
... Super 64B A12B Math REAP BF16256K / 128.6 GB18141
Kenosistron Bf16256K / 199.6 GB290
...20B A12B BF16 REAP 50pct Draft256K / 128.5 GB3186
...n 3 Super 120B A12B BF16 MTPv2256K / 5.9 GB56
Note: green Score (e.g. "73.2") means that the model is better than unsloth/NVIDIA-Nemotron-3-Super-120B-A12B.