Written by Jakub Rusinowski · Last updated September 11, 2026
Model library → MiniCPM-V → MiniCPM-V 4.5
The 8B that made the series' reputation: an average of 77.0 on OpenCompass, which put it ahead of GPT-4o-latest and Gemini 2.0 Pro and made it the strongest multimodal model under 30B at release. Built on Qwen3-8B with a SigLIP2-400M vision encoder, 32K context, 5.6 GB at Q4_K_M. Code is Apache 2.0; the weights carry OpenBMB's own community licence, which permits free commercial use after registration.
MiniCPM-V 4.5 needs about 6 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 | 8 Billion |
| Context window | 32,768 |
| Architecture | SigLIP2-400M encoder + Qwen3-8B decoder |
| Provider | OpenBMB |
| Licence | MiniCPM Model License |
| Specified at | Q4_K_M |
| System RAM | 16 GB |
| Record updated | 2026-09-11 |
MiniCPM Model License — commercial use permitted. Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.
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 | 2.6 GB | 3.4 GB | ~154 tok/s (est.) | Fits comfortably |
| Q3_K_M | 3.4 GB | 4.2 GB | ~133 tok/s (est.) | Fits comfortably |
| Q4_K_M | 4.8 GB | 5.6 GB | ~107 tok/s (est.) | Fits comfortably |
| Q5_K_M | 5.7 GB | 6.5 GB | ~95 tok/s (est.) | Fits comfortably |
| Q6_K | 6.6 GB | 7.4 GB | ~86 tok/s (est.) | Fits comfortably |
| Q8_0 | 8.5 GB | 9.3 GB | ~70 tok/s (est.) | Fits comfortably |
| F16 | 16.0 GB | 16.8 GB | ~41 tok/s (est.) | Fits comfortably |
Want the memory numbers alone, at every quantization level and your own context length? Use the MiniCPM-V 4.5 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 4.5 is the Intel Arc B570 (10 GB).
Install Ollama, then run:
ollama run minicpm-v4.5Weights on Hugging Face: openbmb/MiniCPM-V-4_5.
Quality scores as published by the model's authors or an independent evaluator — not throughput, and not measured by us.
| Benchmark | Score | Provenance |
|---|---|---|
| OpenCompass | 77 / 100 pts | vendor-claimed · https://huggingface.co/openbmb/MiniCPM-V-4_5 |
Best for: multimodal, vision, document analysis, consumer gpu.
← All MiniCPM-V models | VRAM calculator | Check your own hardware