MiniCPM V 4.6 needs about 1.4 GB of memory at Q4_K_M (a 0.5 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 3 quants, from 0.5 GB (Q4_K_M) to 1.5 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 1.6 GB.
MiniCPM V 4.6 – a pocket‑sized multimodal model for image and video understanding
MiniCPM V 4.6 is a 0.752 billion‑parameter model built on Qwen3.5‑0.8B and SigLIP2‑400M. It supports vision tasks and runs efficiently on mobile devices. The model is released under the apache‑2.0 license.
Good at
- Efficient image and video understanding
- Low‑token cost for vision‑language tasks
- Supports mixed 4x/16x visual token compression
Limits
- No text‑only inference without vision
- Requires a GPU with at least 8 GB VRAM for full performance
- No support for large context beyond 262144 tokens
Best for
- Mobile vision‑language apps
- Edge deployment on iOS/Android/HarmonyOS
- Rapid prototyping with llama.cpp or Ollama
Who it’s for. Developers building on‑device vision‑language applications who need a lightweight model that still delivers strong multimodal performance. The model is suitable for mobile and embedded systems with limited GPU memory.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 0.5 GB | 1.4 GB | Fits | Get file → | |
| Q8_0 | 0.8 GB | 1.7 GB | Fits | Get file → | |
| BF16 | 1.5 GB | 2.4 GB | Fits | Get file → |
For 8 GB VRAM use Q4_K_M (0.53 GB file, 1.4 GB memory at 8k context). If more memory is available, Q8_0 (0.81 GB file, 1.7 GB memory) offers higher precision. BF16 (1.52 GB file, 2.4 GB memory) is best for the highest accuracy on GPUs with 12 GB+ VRAM.
What GPU do you need for MiniCPM V 4.6?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is BF16 (1.5 GB file, about 2.4 GB in use).
Which MiniCPM V 4.6 quant should you choose?
Q4_K_M is the usual starting point for MiniCPM V 4.6: it balances file size and quality well. Here it is a 0.5 GB file.
The Q8_0 file is 0.8 GB, about 1.5 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about MiniCPM V 4.6
How much VRAM does MiniCPM V 4.6 need?
About 1.4 GB at Q4_K_M with an 8K context window. That is the 0.5 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run MiniCPM V 4.6 on an 8 GB GPU?
Yes, with BF16: the 1.5 GB file needs about 2.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of MiniCPM V 4.6 GGUF?
It depends on the quant: Q4_K_M is 0.5 GB, Q8_0 is 0.8 GB, BF16 is 1.5 GB.
What context length does MiniCPM V 4.6 support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.4 GB at 8K and 1.6 GB at 32K.
Where can I download MiniCPM V 4.6 GGUF?
The files are on Hugging Face in the ggml-org/MiniCPM-V-4.6-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does MiniCPM V 4.6 use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run MiniCPM V 4.6 on a GPU with 8 GB VRAM?
Yes, the Q4_K_M quantized model requires about 1.4 GB of VRAM at 8k context, which fits comfortably on an 8 GB GPU.
Does MiniCPM V 4.6 support text‑only inference?
No, the model is designed for multimodal tasks and requires vision input to function.
What license applies to MiniCPM V 4.6?
The model is released under the apache‑2.0 license, allowing free use and modification.