Written by Jakub Rusinowski · Last updated September 11, 2026
OpenBMB's pocket-sized multimodal line, backed by ModelBest and Tsinghua University, built around one goal: image and video understanding that actually runs on a phone. Version 4.5 is the 8B that beat much larger proprietary models on OpenCompass; version 4.6 is the 1.3B successor that cut visual-encoding FLOPs by more than half and ships on iOS, Android and HarmonyOS.
| Licence | What it permits | Applies to |
|---|---|---|
Apache-2.0 | Commercial use permitted Commercial use permitted. No usage restrictions beyond attribution. | MiniCPM-V 4.6 |
MiniCPM Model License | Commercial use permitted Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms. | MiniCPM-V 4.5 |
| MiniCPM-V 4.6 | Min 2 GB VRAM · Q4_K_M · 262,144 ctx · ollama run minicpm-v4.6 |
| MiniCPM-V 4.5 | Min 6 GB VRAM · Q4_K_M · 32,768 ctx · ollama run minicpm-v4.5 |
The cheapest GPU that runs MiniCPM-V locally (min 2 GB VRAM) is the Intel Arc B570 (10 GB).
Install Ollama then run: ollama run minicpm-v4.6
Minimum VRAM: 2 GB. For best results use Q4_K_M quantization.
MiniCPM-V needs about 2 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: MiniCPM-V 4.6 (2 GB, Q4_K_M); MiniCPM-V 4.5 (6 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Yes — MiniCPM-V runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4_K_M.
Q4_K_M is the best balance of quality and VRAM for MiniCPM-V in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
Install Ollama, then run: ollama run minicpm-v4.6. This downloads MiniCPM-V and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.