EnlistedGhost/Ministral-3-14B-Instruct-2512-GGUF

------------------------------------------------
- Model Details and Specifications: -
------------------------------------------------

Ministral-3 14B Instruct 2512 (GGUF)

This release contains:
Llama.cpp and Ollama compatible GGUF converted and Quantized model files (Compatible with both Ollama, and Llama.cpp)

Quantized GGUF version of:

• Ministral-3-14B-Instruct-2512-BF16
  (by MistralAI)

Original Model Link:

• mistralai/Ministral-3-14B-Instruct-2512-BF16

Description:
This release includes GGUF (Ollama + Llama.cpp - compatible) model files and three working multi-modal projector(s) (mmproj) files for the Vision Projector; offering full capabilities in Ollama or Llama.cpp.

What's New?
As of February 2026 all available GGUF (both F16 and K-Quants) files have been completely overhauled in order to offer the community what I was able to come up with as best size vs quality possible for this model from MistralAI (Thank you MistralAI for your release!). This was done by first taking the BF16 model's safetensors and up-casting them to F32, from there the model was quantized down to the currently available files you find here. Beyond the F32-->Quant conversion that is a new methodology; These GGUF F16+Quantize files also include a completely custom tokenizer chat template.

Why Up-Cast to F32 from BF16?
Some will argue or perhaps be curious as to why the model was up-cast from BF16 to F32... This is because although it is true that there is no direct quality benefit from casting a BF16 release to a F32 bit file size, there is however a benefit of less quality loss when quantizing from an up-cast BF16 to F32 as the F32 has in a sense more "scratchpad" available mathematically and Quants derrived from an F32 inherently degrade less per-Quant compared to a direct F16 or BF16 conversion. (It's worth a Google if anyone is ever curious!)

What is the "Custom Tokenizer Chat Template?"
As apposed to the standard "Chat Template" made available by MistralAI - this release of GGUF converted and quantized files offer a totally unique and custom Tokenizer Chat Template in order to provide: Smoother, Faster, Efficient, and Sophisticated interaction/inference with the model. This template strips away all of the "fluff" or genuinely unneeded logic from the JINJA Chat/Tokenizer Template - allowing anyone who uses the model for inferencing can enjoy a significant improvement in speed, quality and context adherence without sacrificing any aspect of the initial release by MistralAI.

For reference - here is the new JINJA Tokenizer Chat-Template:
(This template features a sliding context window of EIGHT (8) interactions, which may be adjusted per-individual requirements simply by altering the third line of this template from the number nine (9) to either higher or lower numerical values to increase or decrease the sliding context window)

{{- $remMessage := false }}
{{- range $index, $_ := .Messages }}
{{- if lt (len (slice $.Messages $index)) 9 }}
{{- $remMessage = true }}
{{- end }}
{{- if eq .Role "system" }}[SYSTEM_PROMPT]{{ .Content }}[/SYSTEM_PROMPT]{{- end }}
{{- if $remMessage }}
{{- if eq .Role "user" }}
[INST]{{ .Content }}[/INST]
{{- else if eq .Role "assistant" }}
{{- if .Content}}{{ .Content }}{{- end }}
{{- end }}
{{- end }}
{{- end }}

No modifications, edits, or configurations are required to use this model with Ollama, it works natively! Both Vision and Text work with Ollama. (^.^)

Coming Soon!!!
Check back occasionally - as a automated installer/configure Python-3 script is making its way to all of my releases! This allows anyone who is interested in using these models a hassle-free and stress-free experience where the Python-3 script takes care of setting up the model for Ollama (specifically for Ollama, other software optimizations coming later). It is highly recommended to use the Ollama "create" command along with the supplied ".modelfile" to ensure proper configuration for anyone who wishes to get the most out of this particular release. Though, the Python-3 automated installer/configuration tool will handle such aspects if it is chosen to be used.

Happy Inferencing!
-- Jon Z (EnlistedGhost)

---

Model Updates (As of: February 27th, 2026)

• Updated: All GGUF model file(s) with significantly altered/reworked GGUF conversion(s) and Quant file(s)
  Final Quantized and full-F16 modelfiles are being uploaded!!! - Check back for I-Matrix quant files and other static K-Quants if you do not see your desired edition (They are being uploaded, thank you for your patience!)

---

-------------------------------------------------------------
- GGUF Conversion and Quantization Details: -
-------------------------------------------------------------

Software used to convert Safetensors to GGUF:

• llama.cpp

Software used to create Quantized GGUF Files:

• llama.cpp

Specific GitHub Commit Point:

• b8118

Converted to GGUF and Quantized by:

• EnlistedGhost

---

--------------------------
---- Original Info ----
--------------------------

(Crossposted from the link in the above section: "Model Details"):

Ministral 3 14B Instruct 2512 BF16

The largest model in the Ministral 3 family, Ministral 3 14B offers frontier capabilities and performance comparable to its larger Mistral Small 3.2 24B counterpart. A powerful and efficient language model with vision capabilities.

This model is the instruct post-trained version, fine-tuned for instruction tasks, making it ideal for chat and instruction based use cases.

The Ministral 3 family is designed for edge deployment, capable of running on a wide range of hardware. Ministral 3 14B can even be deployed locally, capable of fitting in 32GB of VRAM in BF16, and less than 24GB of RAM/VRAM when quantized.

We provide a no-loss FP8 version here, you can find other formats and quantizations in the Ministral 3 - Additional Checkpoints collection.

Key Features

Ministral 3 14B consists of two main architectural components:

• 13.5B Language Model
• 0.4B Vision Encoder

The Ministral 3 14B Instruct model offers the following capabilities:

• Vision: Enables the model to analyze images and provide insights based on visual content, in addition to text.
• Multilingual: Supports dozens of languages, including English, French, Spanish, German, Italian, Portuguese, Dutch, Chinese, Japanese, Korean, Arabic.
• System Prompt: Maintains strong adherence and support for system prompts.
• Agentic: Offers best-in-class agentic capabilities with native function calling and JSON outputting.
• Edge-Optimized: Delivers best-in-class performance at a small scale, deployable anywhere.
• Apache 2.0 License: Open-source license allowing usage and modification for both commercial and non-commercial purposes.
• Large Context Window: Supports a 256k context window.

Use Cases

Private AI deployments where advanced capabilities meet practical hardware constraints:

• Private/custom chat and AI assistant deployments in constrained environments
• Advanced local agentic use cases
• Fine-tuning and specialization
• And more...

Bringing advanced AI capabilities to most environments.

Ministral 3 Family

Model Name	Type	Precision	Link
Ministral 3 3B Base 2512	Base pre-trained	BF16	Hugging Face
Ministral 3 3B Instruct 2512	Instruct post-trained	BF16	Hugging Face
Ministral 3 3B Reasoning 2512	Reasoning capable	BF16	Hugging Face
Ministral 3 8B Base 2512	Base pre-trained	BF16	Hugging Face
Ministral 3 8B Instruct 2512	Instruct post-trained	BF16	Hugging Face
Ministral 3 8B Reasoning 2512	Reasoning capable	BF16	Hugging Face
Ministral 3 14B Base 2512	Base pre-trained	BF16	Hugging Face
Ministral 3 14B Instruct 2512	Instruct post-trained	BF16	Hugging Face
Ministral 3 14B Reasoning 2512	Reasoning capable	BF16	Hugging Face

Other formats available here.

Benchmark Results

We compare Ministral 3 to similar sized models.

Reasoning

Model	AIME25	AIME24	GPQA Diamond	LiveCodeBench
Ministral 3 14B	0.850	0.898	0.712	0.646
Qwen3-14B (Thinking)	0.737	0.837	0.663	0.593
Ministral 3 8B	0.787	0.860	0.668	0.616
Qwen3-VL-8B-Thinking	0.798	0.860	0.671	0.580
Ministral 3 3B	0.721	0.775	0.534	0.548
Qwen3-VL-4B-Thinking	0.697	0.729	0.601	0.513

Instruct

Model	Arena Hard	WildBench	MATH Maj@1	MM MTBench
Ministral 3 14B	0.551	68.5	0.904	8.49
Qwen3 14B (Non-Thinking)	0.427	65.1	0.870	NOT MULTIMODAL
Gemma3-12B-Instruct	0.436	63.2	0.854	6.70
Ministral 3 8B	0.509	66.8	0.876	8.08
Qwen3-VL-8B-Instruct	0.528	66.3	0.946	8.00
Ministral 3 3B	0.305	56.8	0.830	7.83
Qwen3-VL-4B-Instruct	0.438	56.8	0.900	8.01
Qwen3-VL-2B-Instruct	0.163	42.2	0.786	6.36
Gemma3-4B-Instruct	0.318	49.1	0.759	5.23

Base

Model	Multilingual MMLU	MATH CoT 2-Shot	AGIEval 5-shot	MMLU Redux 5-shot	MMLU 5-shot	TriviaQA 5-shot
Ministral 3 14B	0.742	0.676	0.648	0.820	0.794	0.749
Qwen3 14B Base	0.754	0.620	0.661	0.837	0.804	0.703
Gemma 3 12B Base	0.690	0.487	0.587	0.766	0.745	0.788
Ministral 3 8B	0.706	0.626	0.591	0.793	0.761	0.681
Qwen 3 8B Base	0.700	0.576	0.596	0.794	0.760	0.639
Ministral 3 3B	0.652	0.601	0.511	0.735	0.707	0.592
Qwen 3 4B Base	0.677	0.405	0.570	0.759	0.713	0.530
Gemma 3 4B Base	0.516	0.294	0.430	0.626	0.589	0.640

License

This model is licensed under the Apache 2.0 License.

You must not use this model in a manner that infringes, misappropriates, or otherwise violates any third party’s rights, including intellectual property rights.