Skip to content
NEWLive updates feed from Hugging Face, arXiv and GitHub  See what changed →

nvidia · gemma4 · based on google/gemma-4-26B-A4B-it

Gemma-4-26B-A4B-NVFP4

Gemma-4-26B-A4B-NVFP4 by nvidia: a 14.4B-parameter mixture-of-experts open-weight model under the apache-2.0 license and vision.

apache-2.0Mixture of experts262K contextVisionAuto-imported · not yet reviewed

1,807,367 downloads · 148 likes · Model card · synced 2026-09-19

Specifications

Parameters
14.4B
Architecture
moe · gemma4
Layers
30
Hidden size
2,816
KV heads · head dim
8 · 256
Vocabulary
262,144
Native dtype
Context window
262,144 tokens
Released
2026-05-01
Last modified on Hub
2026-05-11

Features & licensing

License
apache-2.0
Commercial use
Yes
Modalities
text, vision
Tool / function calling
No
Reasoning mode
No
Languages
EN
Quantised variants on Hub
GGUF
Pipeline
text-generation
Model Optimizersafetensorsgemma4nvidiaModelOptquantizedNVFP4nvfp4gemma4-26b-A4B-ittext-generationconversational8-bitmodelopt

Task fit

Editorial scores (0–100) used by the recommendation engine.

  • Chat70
  • RAG / Q&A70
  • Code70
  • Summarisation70
  • Extraction70
  • Agents70

Hardware to run Gemma-4-26B-A4B-NVFP4

Weights need about 32 GB at FP16, 17 GB at INT8 and 10 GB at INT4. The KV cache adds roughly 246 MB per 1,000 tokens per request. For a reference workload of 5 requests per second with 1,500 input and 300 output tokens, total GPU memory is around 15.7 GB, which fits on 1 × NVIDIA A100 (80 GB).

GPU configurations at the reference workload
ConfigurationVRAMUtilisationEst. first tokenCloud / month
1 × NVIDIA L4 (24 GB)24 GB65 %~0.5 s$584
1 × NVIDIA L40S (48 GB)48 GB33 %~0.2 s$1,387
1 × NVIDIA A100 (80 GB)recommended80 GB20 %~0.1 s$2,336
1 × NVIDIA H100 (80 GB)80 GB20 %~0.0 s$3,285

Estimates only. Run an assessment for your own traffic.

Auto-imported from Hugging Face. Architecture, license and feature data are synced from the model repository; the task-fit scores below are provisional defaults until a curator reviews them.