---

license: apache-2.0

base_model: mistralai/Ministral-3-8B-Instruct-2512

tags:

- gguf

- nobodywho

- tool-calling

- vision

- mistral

- ministral

pipeline_tag: image-text-to-text

library_name: gguf

---

NobodyWho/Mistral_Ministral-3-8B-Instruct-GGUF

Overview

GGUF quantization of Mistral AI's Ministral 3 8B Instruct (Ministral-3-8B-Instruct-2512),

prepared for NobodyWho: it works with NobodyWho out of

the box, with Mistral's recommended sampling metadata embedded in every quant, and is verified

with NobodyWho's test suite. Ministral 3 8B is a recent,

edge-focused multimodal model (8.4B language model + 0.4B vision encoder) with best-in-class

agentic tool calling, released under the Apache 2.0 license.

Model Capabilities

• Text generation — instruction-following chat
• Tool calling — native function calling with JSON output and grammar constraints
• Vision — image understanding via the companion mmproj-BF16.gguf projection model
• Long context — 256k tokens
• Multilingual — dozens of languages

Available Quantizations

| File | Approach | Tool-calling tests |

|------|----------|--------------------|

| Ministral-3-8B-Instruct-2512-BF16-vendor-sampling.gguf | Vendor sampling injected | 14/14 |

| Ministral-3-8B-Instruct-2512-Q8_0-vendor-sampling.gguf | Vendor sampling injected | 14/14 |

| Ministral-3-8B-Instruct-2512-Q4_K_M-vendor-sampling.gguf | Vendor sampling injected | 14/14 |

| mmproj-BF16.gguf | Vision projection (use with any of the above) | — |

> Verified with NobodyWho's tool-calling suite across BF16 / Q8_0 / Q4_K_M (14/14 each, June

> 2026); vision and multilingual verified. The upstream GGUF has no general.sampling.* metadata, so the

> -vendor-sampling files embed Mistral's recommended sampler (see INJECTION.md).

Quick Start

Using the NobodyWho library:

from nobodywho import Chat

chat = Chat("huggingface:NobodyWho/Mistral_Ministral-3-8B-Instruct-GGUF/Ministral-3-8B-Instruct-2512-Q4_K_M-vendor-sampling.gguf")
response = chat.ask("What is the capital of Denmark?").completed()
print(response)  # The capital of Denmark is Copenhagen.

Vision

from nobodywho import Model, Chat, Prompt, Image, Text

model = Model(
    "huggingface:NobodyWho/Mistral_Ministral-3-8B-Instruct-GGUF/Ministral-3-8B-Instruct-2512-Q4_K_M-vendor-sampling.gguf",
    projection_model_path="huggingface:NobodyWho/Mistral_Ministral-3-8B-Instruct-GGUF/mmproj-BF16.gguf",
)
chat = Chat(model=model, system_prompt="You are a helpful assistant.")
response = chat.ask(Prompt([
    Text("What is in this image?"),
    Image("./photo.png"),
])).completed()
print(response)

llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
    repo_id="NobodyWho/Mistral_Ministral-3-8B-Instruct-GGUF",
    filename="Ministral-3-8B-Instruct-2512-Q4_K_M-vendor-sampling.gguf",
)

Model Specifications

• Parameters: 8.4B language model + 0.4B vision encoder
• Context length: 262,144 tokens (256K)
• License: Apache 2.0
• Base model: mistralai/Ministral-3-8B-Instruct-2512
• Architecture: mistral3 (vision-capable)

Licensing / Credits

Licensed under Apache 2.0 (unchanged from upstream). All model credit belongs to Mistral AI.

GGUF quantizations provided by unsloth.