Autor: Jakub Rusinowski · Ostatnia aktualizacja: 11 września 2026
Model library → MiniCPM-V → MiniCPM-V 4.6
1.3 billion parameters — about 1.6 GB at Q4_K_M including cache and overhead, which is genuinely phone-sized. A SigLIP2-400M vision encoder on a Qwen3.5-0.8B backbone, following LLaVA-UHD v4, with mixed 4x/16x visual token compression that cuts visual encoding FLOPs by more than 50%. Takes text, images and video (up to 128 frames) across a 262,144-token context. Apache 2.0, with day-one llama.cpp, Ollama, vLLM and SGLang support.
MiniCPM-V 4.6 needs about 2 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
| Parameters | 1.3 Billion |
| Context window | 262,144 |
| Architecture | SigLIP2-400M encoder + Qwen3.5-0.8B decoder |
| Provider | OpenBMB |
| Licence | Apache 2.0 |
| Specified at | Q4_K_M |
| System RAM | 8 GB |
| Record updated | 2026-09-11 |
Apache-2.0 — commercial use permitted. Commercial use permitted. No usage restrictions beyond attribution.
Modelled on a reference NVIDIA RTX 4090 (24 GB), with no KV cache (this record has no published architecture). Speed figures are ESTIMATES from the memory-bandwidth roofline described on the methodology page, not benchmarks we ran — rows marked measured come from published or reader-submitted runs. VRAM here includes the KV cache, so it reads higher than the headline figure above, which does not.
| Quant | Weights | VRAM needed | Est. speed | Fit on 24 GB |
|---|---|---|---|---|
| Q2_K | 0.4 GB | 1.2 GB | ~309 tok/s (est.) | Fits comfortably |
| Q3_K_M | 0.6 GB | 1.4 GB | ~294 tok/s (est.) | Fits comfortably |
| Q4_K_M | 0.8 GB | 1.6 GB | ~270 tok/s (est.) | Fits comfortably |
| Q5_K_M | 0.9 GB | 1.7 GB | ~257 tok/s (est.) | Fits comfortably |
| Q6_K | 1.1 GB | 1.9 GB | ~245 tok/s (est.) | Fits comfortably |
| Q8_0 | 1.4 GB | 2.2 GB | ~222 tok/s (est.) | Fits comfortably |
| F16 | 2.6 GB | 3.4 GB | ~164 tok/s (est.) | Fits comfortably |
Want the memory numbers alone, at every quantization level and your own context length? Use the MiniCPM-V 4.6 VRAM calculator.
or compare on Vast.ai from $0.35/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
The cheapest catalogued GPU that runs MiniCPM-V 4.6 is the Intel Arc B570 (10 GB).
Install Ollama, then run:
ollama run minicpm-v4.6Weights on Hugging Face: openbmb/MiniCPM-V-4.6.
Best for: multimodal, vision, edge devices, mobile.
← All MiniCPM-V models | VRAM calculator | Check your own hardware