Model - Pixtral 12B (GGUF) (Ollama Patched)

Description: This is an Ollama patched version of Pixtral-12B (GGUF) with working projector (mmproj) files. No modifications, edits, or configuration is required to use this model with Ollama, it works natively! Both Vision and Text work with Ollama. (^.^)

IMPORTANT NOTICE as of (GMT-8) 01:04 November 19th, 2025: After significant amounts of self-testing of the supplied Quantized GGUF files made available within this release I have been pleasantly surprised to find that so far ALL of the files have the AI performing quite admirably despite Huggingface's systems here only merging in the Q8_0 (8-bit quantized) Vision Projector - with no automated option to choose a higher quality beyond assuming that you already know how to use Ollama's "create" command and so on.

I will be updating this release with both:

• Instructions within this model page and linked documentation on how to easily "create" a customized/tailored Multi-Modal model that you may speak/inference with utilizing Ollama (or llama.cpp as well).
• Automated scripts via bash (Linux users) and command-line (Windows users) options that will allow anyone to simply download the small script file and then merely run (click on) the script which will interactively guide you through a user-friendly setup process - including merging the higher quality Vision Projector options.

In addition to the above mentioned updates:

I am currently releasing GGUF Ollama-ready (Natively Ollama compatible) alternate versions of Pixtral-12B. In case some of you are not aware of this fact: there are currently two other Officially released versions of Pixtral-12B. These are supplied directly from MistralAI themselves and include the Pixtral-12B-2409 official release and the the Pixtral-12B-Base-2409 models that, unlike the release that these files were converted and quantized from, ARE NOT a community release. These two other versions differ in quality and capability, although slightly, from the originating "mistral-community" release that this GGUF release (and MOST other releases or derivatives) is based upon. This is because MistralAI's two releases with the "2409" label are not user-friendly to setup or use. Thankfully I have humbly managed to have them successfully loaded, converted to GGUF and now am in the process of creating quality Quantized GGUF versions of specifically for use with Ollama/Llama.cpp.

Be sure to check back occasionally as I expect to make these more rarely seen Pixtral-12B releases from MistralAI made available within the next couple days! (Rarely seen - to clarify what I meant was that they are extremely uncommong to find GGUF versions of, and not many versions even that are not GGUF)

Happy Inferencing! -- Jon Z (EnlistedGhost)

Model Updates (As of: November 19th, 2025)

Recently finished updates:

• Partially Updated: ModelCard (this page)

Currently in-progress updates:

• Update: ModelCard (this page needs more setup guides and information...)

How to run this Model using Ollama

You can run this model by using the "ollama run" command. Simply copy & paste one of the commands from the list below into your console, terminal or power-shell window.

mmproj (Vision Projector) Files

Intended Use

Same as original:

• [mistralai]

Out-of-Scope Use

Same as original:

• [mistralai]

Bias, Risks, and Limitations

Same as original:

• [mistralai]

Training Details

Training sets and data are from:

• [mistralai]
• [mistralai-community] (This is a direct off-shoot/decendant of the above mentioned models)

Evaluation

• This model has NOT been evaluated in any form, scope or type of method.
• !!! USE AT YOUR OWN RISK !!!
• !!! NO WARRANTY IS PROVIDED OF ANY KIND !!!

Citation (Original Paper)

[MistalAI Pixtral-12B Original Paper]

Detailed Release Information

• Originally Developed by: [mistralai]
• Further Developed by: [mistralai-community]
• MMPROJ (Vision) Quantized by: [EnlistedGhost]
• Model Quantized for GGUF by: [EnlistedGhost]
• Modified for Ollama by: [EnlistedGhost]
• Released on Huggingface by: [EnlistedGhost]
• Model type & format: [Quantized/GGUF]
• License type: [Apache-2.0]

Attributions (Credits)

A big thank-you is extended to the below credited sources! These contributions are what made this release possible!

Important Notice: This is NOT a copy/paste release, I have created unique Quantized files that were then altered further to properly work with Ollama software. This resulted in the first publicly available Pixtral-12B model that natively runs on Ollama.

• [ggml-org] (Resources no longer utilized)
• [mrmadermacher] (Resources no longer utilized)
• [bartowski] (Still using low iMatrix Quant files: IQ2_M and IQ2_S)
• [Ollama developers]

Model Card Authors and Contact

[EnlistedGhost]