ZoomGain-9B-GGUF Vision-Language Model: 8GB+ VRAM

ZoomGain-9B-GGUF Vision-Language Model: 8GB+ VRAM

At a Glance

Item Value
Repository inclusionAI/ZoomGain-9B-GGUF
Family guide ZoomGain guide (3 articles)
Publisher guide inclusionAI (Ant Group): models and licenses
Published 2026-11-11
Formats GGUF
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

Overview

inclusionAI has released model weights converted into the GGUF format for the vision-language model “ZoomGain-9B", available as “inclusionAI/ZoomGain-9B-GGUF“.

ZoomGain-9B-GGUF is an open-weight model converted into the GGUF format using llama.cpp. It supports multimodal image-text-to-text (vision-language) tasks, accepting both image and text inputs to generate response text. Equipped with multimodal conversational capabilities based on the Qwen3.5 architecture, it is primarily provided for engineers and developers who want to run vision-language models on local PCs or servers.

In this repository, alongside the language model weights themselves, an mmproj file containing the vision encoder and projector—which handle image analysis and projection into the language space—is distributed simultaneously. When performing image-input processing during local inference, it is necessary to prepare both the language model’s GGUF file and its corresponding mmproj file, and load them into the execution engine.

Available precision variants include FP16 for full precision, as well as quantized versions Q8_0 and Q4_K_M. These precision options are provided for both the language model and the mmproj file, allowing combinations to be chosen according to the memory limitations and computational resources of the execution environment. Note that the model card from the publisher cautions that applying quantization processing may affect the output accuracy of the model. Additionally, execution assumes the use of a build of llama.cpp that supports the multimodal features of Qwen3.5.

Specifications

Specifications explicitly stated in the model card and related materials are as follows:

  • Language model available precisions: FP16, Q8_0, Q4_K_M
  • Vision encoder and projector (mmproj) available precisions: FP16, Q8_0, Q4_K_M
  • Supported pipeline: image-text-to-text / vision-language / conversational
  • Distribution format: GGUF

Strengths and Use Cases

ZoomGain-9B-GGUF excels at vision-language tasks that simultaneously understand input image content and text instructions to output response text. Specifically, it is designed for use in multimodal conversations such as detailed image descriptions, situation explanations, and answering questions about images.

Featuring image-text processing capabilities based on the Qwen3.5 architecture and converted into the GGUF format, it is well-suited for running inference in local environments using llama.cpp. It can be utilized when running vision-language models on a personal PC or local server without sending images to external cloud services, or for building local web chat systems combined with conversational UIs.

Note that this model consists of weights obtained by quantizing and converting the original ZoomGain-9B model. Users should be aware that quantization processing may potentially impact output accuracy.

Hardware Requirements

Estimated requirements (calculated by Local Model Watch)

Your VRAM Quantization File size Est. memory needed
8GB (RTX 4060 / 3060 Ti, etc.) Q4_K_M 5.2GB 6.3GB
12GB (RTX 4070 / 3060 12GB, etc.) Q8_0 8.9GB 10.6GB
24GB (RTX 4090 / 3090, etc.) F16 16.7GB 20.0GB

Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model’s authors. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.

Can You Run It Locally?

Runs in Ollama, LM Studio and llama.cpp as-is.

It is distributed in GGUF, so no conversion is needed.

Compiled by this site’s code from the published formats, converted builds we have found, and each engine’s own model registry. “Not found" means we have not seen such a build, not that none exists.

Our Own Measurements

Measurements We Did Not Take

This model does not state a license, so we cannot confirm permission to use it. Because this site carries advertising, we did not run the model (no CPU run, answers, quantization comparison, conversion or generation).

How to Get It

From the ZoomGain-9B-GGUF repository, the GGUF format language model weights and an mmproj file containing the vision encoder and projector are distributed. Obtaining and running inference requires a build of llama.cpp that supports the multimodal features of Qwen3.5.

When performing inference involving image inputs, specify and load the language model file (-m) alongside the mmproj file (--mmproj) of the same or corresponding precision simultaneously.

When running inference in a CLI environment, use llama-mtmd-cli as follows:

llama-mtmd-cli \
  -m ZoomGain-9B-Q8_0.gguf \
  --mmproj mmproj-ZoomGain-9B-Q8_0.gguf \
  --image /path/to/image.jpg \
  -p "Describe this image." \
  --temp 0 -n 2048

To operate it as a web chat or API server, start the server using llama-server:

llama-server \
  -m ZoomGain-9B-Q8_0.gguf \
  --mmproj mmproj-ZoomGain-9B-Q8_0.gguf \
  --host 127.0.0.1 --port 8080

By replacing Q8_0 in the commands with F16 or Q4_K_M, you can select and run the file corresponding to your desired precision.

Related Articles

What to Read Next

Sources