MiniCPM-V 4.6 — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
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.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Weights plus overhead only: this model's architecture is unpublished, so no KV cache is added.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 1.2 GB | ~309 tok/s | Fits | |
| Q3_K_M 3.41 bpw | 1.4 GB | ~294 tok/s | Fits | |
| Q4_K_M 4.83 bpw | 1.6 GB | ~270 tok/s | Fits | |
| Q5_K_M 5.67 bpw | 1.7 GB | ~257 tok/s | Fits | |
| Q6_K 6.56 bpw | 1.9 GB | ~245 tok/s | Fits | |
| Q8_0 8.50 bpw | 2.2 GB | ~222 tok/s | Fits | |
| F16 16.00 bpw | 3.4 GB | ~164 tok/s | Fits |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. MiniCPM-V 4.6 VRAM calculator →
Get MiniCPM-V 4.6 running
The cheapest catalogued GPU that runs MiniCPM-V 4.6 is the Intel Arc B570 (10 GB).
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.
How to run MiniCPM-V 4.6
Install Ollama, then run:
ollama run minicpm-v4.6Specifications
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- Parameters
- 1.3 Billion
- Context window
- 262,144
- Architecture
- SigLIP2-400M encoder + Qwen3.5-0.8B decoder
- Provider
- OpenBMB
- Licence
- Apache 2.0
- Practical deployment
- Edge device (phone, SBC)
- Specified at
- Q4_K_M
- System RAM
- 8 GB
- Record updated
- 2026-09-11
Commercial use permitted. No usage restrictions beyond attribution.
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Can I run MiniCPM-V 4.6 on my GPU?
- MiniCPM-V 4.6 on AMD Radeon RX 9060 XT 8GB
- MiniCPM-V 4.6 on Intel Arc B570
- MiniCPM-V 4.6 on Intel Arc B580
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 3060 (12GB)
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 3070
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 3070 Ti
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 3080 (10GB)
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 4060
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 4070
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 4070 Super
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 4070 Ti
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 5060
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 5060 Ti 8GB
- MiniCPM-V 4.6 on NVIDIA GeForce RTX 5070
Other MiniCPM-V sizes
MiniCPM-V 4.6 — frequently asked questions
How much VRAM does MiniCPM-V 4.6 need?
About 2 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does MiniCPM-V 4.6 run on an RTX 4090 (24 GB)?
Yes. MiniCPM-V 4.6 needs about 2 GB at Q4_K_M, inside a 24 GB card, at an estimated 270 tokens/sec.
How do I run MiniCPM-V 4.6 locally?
Install Ollama and run `ollama run minicpm-v4.6`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.
What other sizes does MiniCPM-V come in?
MiniCPM-V 4.6 (2 GB), MiniCPM-V 4.5 (6 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.