What are the hardware requirements for Granite 4.0 3B Vision?

Granite 4.0 3B Vision requires approximately 8.5 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Granite 4.0 3B Vision and how large is it?

Granite 4.0 3B Vision is developed by ibm-granite, a model with 3b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate Granite 4.0 3B Vision?

Granite 4.0 3B Vision is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Granite 4.0 3B Vision by ibm-granite — VRAM 8.5GB

Name: Granite 4.0 3B Vision
Author: ibm-granite

Granite 4.0 3B Vision is an open-source language model by ibm-granite. Features: 3b LLM, VRAM: 8.5GB, License: apache-2.0, LLM Explorer Score: 0.27.

Arxiv:2208.00385 Arxiv:2404.19205 Arxiv:2406.04334 Arxiv:2412.07626 Arxiv:2502.09927 Arxiv:2512.10888 Arxiv:2603.27064 Conversational Custom code En Endpoints compatible Granite4 vision Image-text-to-text Lora Region:us Safetensors Sharded Tensorflow Vision

Model Card on HF 🤗: https://huggingface.co/ibm-granite/granite-4.0-3b-vision

Granite 4.0 3B Vision Parameters and Internals

LLM Name	Granite 4.0 3B Vision
Repository 🤗	https://huggingface.co/ibm-granite/granite-4.0-3b-vision
Model Size	3b
Required VRAM	8.5 GB
Updated	2026-08-08
Maintainer	ibm-granite
Model Files	0.9 GB 5.0 GB: 1-of-2 3.5 GB: 2-of-2
Supported Languages	en
Model Architecture	AutoModelForMultimodalLM
License	apache-2.0
Is Biased	none
Tokenizer Class	GPT2Tokenizer
Padding Token	<\|pad\|>
PEFT Type	LORA
LoRA Model	Yes
PEFT Target Modules	language_model.layers.20.self_attn.q_proj\|language_model.layers.7.self_attn.q_proj\|32.self_attn.k_proj\|28.self_attn.q_proj\|language_model.layers.14.self_attn.k_proj\|language_model.layers.19.self_attn.v_proj\|language_model.layers.26.self_attn.k_proj\|29.self_attn.q_proj\|35.self_attn.k_proj\|language_model.layers.8.self_attn.v_proj\|34.self_attn.k_proj\|language_model.layers.22.self_attn.q_proj\|29.self_attn.v_proj\|35.self_attn.v_proj\|31.self_attn.q_proj\|language_model.layers.4.self_attn.q_proj\|language_model.layers.18.self_attn.v_proj\|language_model.layers.10.self_attn.v_proj\|27.self_attn.v_proj\|34.self_attn.v_proj\|language_model.layers.25.self_attn.v_proj\|language_model.layers.21.self_attn.v_proj\|language_model.layers.12.self_attn.q_proj\|35.self_attn.q_proj\|38.self_attn.q_proj\|language_model.layers.6.self_attn.q_proj\|language_model.layers.22.self_attn.k_proj\|language_model.layers.25.self_attn.q_proj\|language_model.layers.16.self_attn.q_proj\|language_model.layers.16.self_attn.v_proj\|language_model.layers.23.self_attn.k_proj\|language_model.layers.13.self_attn.q_proj\|36.self_attn.q_proj\|39.self_attn.q_proj\|language_model.layers.2.self_attn.v_proj\|37.self_attn.v_proj\|27.self_attn.q_proj\|language_model.layers.20.self_attn.k_proj\|language_model.layers.1.self_attn.k_proj\|language_model.layers.1.self_attn.q_proj\|37.self_attn.k_proj\|input_linear\|28.self_attn.k_proj\|language_model.layers.12.self_attn.k_proj\|language_model.layers.23.self_attn.v_proj\|27.self_attn.k_proj\|language_model.layers.1.self_attn.v_proj\|language_model.layers.7.self_attn.v_proj\|language_model.layers.6.self_attn.k_proj\|language_model.layers.10.self_attn.k_proj\|39.self_attn.k_proj\|language_model.layers.14.self_attn.v_proj\|language_model.layers.26.self_attn.v_proj\|language_model.layers.15.self_attn.v_proj\|language_model.layers.8.self_attn.k_proj\|language_model.layers.9.self_attn.v_proj\|language_model.layers.16.self_attn.k_proj\|30.self_attn.v_proj\|language_model.layers.3.self_attn.q_proj\|33.self_attn.k_proj\|36.self_attn.k_proj\|o_proj\|33.self_attn.v_proj\|language_model.layers.15.self_attn.k_proj\|language_model.layers.9.self_attn.k_proj\|language_model.layers.3.self_attn.v_proj\|32.self_attn.q_proj\|language_model.layers.6.self_attn.v_proj\|language_model.layers.0.self_attn.k_proj\|language_model.layers.2.self_attn.q_proj\|language_model.layers.9.self_attn.q_proj\|language_model.layers.18.self_attn.k_proj\|38.self_attn.k_proj\|language_model.layers.14.self_attn.q_proj\|language_model.layers.0.self_attn.q_proj\|30.self_attn.k_proj\|language_model.layers.13.self_attn.k_proj\|language_model.layers.11.self_attn.v_proj\|language_model.layers.4.self_attn.v_proj\|31.self_attn.k_proj\|language_model.layers.13.self_attn.v_proj\|language_model.layers.5.self_attn.v_proj\|28.self_attn.v_proj\|language_model.layers.24.self_attn.k_proj\|language_model.layers.23.self_attn.q_proj\|language_model.layers.2.self_attn.k_proj\|language_model.layers.5.self_attn.k_proj\|language_model.layers.19.self_attn.k_proj\|language_model.layers.7.self_attn.k_proj\|language_model.layers.4.self_attn.k_proj\|language_model.layers.19.self_attn.q_proj\|language_model.layers.10.self_attn.q_proj\|language_model.layers.25.self_attn.k_proj\|language_model.layers.18.self_attn.q_proj\|36.self_attn.v_proj\|language_model.layers.21.self_attn.k_proj\|33.self_attn.q_proj\|language_model.layers.24.self_attn.q_proj\|34.self_attn.q_proj\|output_linear\|language_model.layers.5.self_attn.q_proj\|39.self_attn.v_proj\|38.self_attn.v_proj\|language_model.layers.0.self_attn.v_proj\|language_model.layers.17.self_attn.q_proj\|language_model.layers.3.self_attn.k_proj\|29.self_attn.k_proj\|language_model.layers.26.self_attn.q_proj\|language_model.layers.11.self_attn.k_proj\|language_model.layers.24.self_attn.v_proj\|language_model.layers.17.self_attn.v_proj\|37.self_attn.q_proj\|language_model.layers.22.self_attn.v_proj\|language_model.layers.17.self_attn.k_proj\|language_model.layers.21.self_attn.q_proj\|30.self_attn.q_proj\|language_model.layers.11.self_attn.q_proj\|language_model.layers.8.self_attn.q_proj\|language_model.layers.12.self_attn.v_proj\|language_model.layers.20.self_attn.v_proj\|32.self_attn.v_proj\|31.self_attn.v_proj\|language_model.layers.15.self_attn.q_proj
LoRA Alpha	256
LoRA Dropout	0.05
R Param	256
Errors	replace

Rank the Granite 4.0 3B Vision capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Quantized Models of the Granite 4.0 3B Vision

Model	Likes	Downloads	VRAM
Granite 4.0 3B Vision 5bit	1	12	3 GB

Best Alternatives to Granite 4.0 3B Vision

Best Alternatives	Context / RAM	Downloads	Likes
...inistral 3 3B Glm Distillation	256K / 7.7 GB	97	1

Note: green Score (e.g. "73.2") means that the model is better than ibm-granite/granite-4.0-3b-vision.