8-bit Open on Hugging Face
Parameters
3.8B
Weights (est.)
≈ 4.5 GB
Mac memory (est.)
8 GB+
Size is estimated from the parameter count and quantisation. Memory allows a fifth extra for context plus 2 GB for the system; long contexts need more.
Run it
With mlx-vlm on an Apple silicon Mac:
pip install mlx-vlmpython -m mlx_vlm.generate --model mlx-community/Ministral-3-3B-Base-2512-8bit --image photo.jpg --prompt "Describe this image"
Details
- Base model
- mistralai/Ministral-3-3B-Base-2512 (quantized)
- Vendor
- Mistral
- Task
- Vision + text
- Quantisation
- 8-bit
- Licence
- apache-2.0
- Downloads
- 91 in the last 30 days · 425 all time · 0 likes
- Published
- 2026-07-18 · updated 2026-07-18 (3mo ago)
- Languages
- en, fr, es, de, it, pt, nl, zh, ja, ko, ar
- Tags
- mistral3, mistral-common, ministral, ministral-3, vision-language, multimodal, quantized, edge, base-model
Other MLX builds of mistralai/Ministral-3-3B-Base-2512
| Build | Quantisation | Weights | Mac memory | Downloads |
|---|---|---|---|---|
| Ministral-3-3B-Instruct-2512-4bit | 4-bit | ≈ 2.7 GB | 8 GB+ | 20K |
| Ministral-3-3B-Reasoning-2512-4bit | 4-bit | ≈ 2.7 GB | 8 GB+ | 765 |
| Ministral-3-3B-Base-2512-4bit | 4-bit | ≈ 2.8 GB | 8 GB+ | 190 |
| Ministral-3-3B-Instruct-2512-5bit | 5-bit | ≈ 3.2 GB | 8 GB+ | 61 |
| Ministral-3-3B-Reasoning-2512-5bit | 5-bit | ≈ 3.2 GB | 8 GB+ | 36 |
| Ministral-3-3B-Instruct-2512-6bit | 6-bit | ≈ 3.6 GB | 8 GB+ | 70 |
| Ministral-3-3B-Reasoning-2512-6bit | 6-bit | ≈ 3.6 GB | 8 GB+ | 34 |
| Ministral-3-3B-Base-2512-6bit | 6-bit | ≈ 3.6 GB | 8 GB+ | 64 |
| Ministral-3-3B-Instruct-2512-8bit | 8-bit | ≈ 4.5 GB | 8 GB+ | 209 |
| Ministral-3-3B-Reasoning-2512-8bit | 8-bit | ≈ 4.5 GB | 8 GB+ | 96 |
| Ministral-3-3B-Base-2512-8bit (this one) | 8-bit | ≈ 4.5 GB | 8 GB+ | 91 |
| Ministral-3-3B-Instruct-2512-bf16 | bf16 | ≈ 7.7 GB | 16 GB+ | 113 |
| Ministral-3-3B-Reasoning-2512-bf16 | bf16 | ≈ 7.7 GB | 16 GB+ | 112 |
| Ministral-3-3B-Base-2512-bf16 | bf16 | ≈ 7.7 GB | 16 GB+ | 48 |