llmrun
OpenBMBMiniCPMMiniCPMV

MiniCPM V 4 — Hardware Requirements & GPU Compatibility

Vision

MiniCPM-V 4.0 is OpenBMB's multimodal model with 4.1 billion parameters, built on SigLIP2-400M and MiniCPM4-3B for on-device use. It handles single-image, multi-image, and video understanding; the card reports an OpenCompass average of 69.0, beating MiniCPM-V 2.6 (which has 8.1 billion parameters) as well as Qwen2.5-VL-3B-Instruct. The card also cites under two seconds to first token and more than 17 tokens per second on an iPhone 16 Pro Max. At this size, it runs on modest consumer GPUs, laptops, and even phones once quantized. Context length reaches 32,768 tokens. Released under the Apache 2.0 license for commercial and research use, it appeared in July 2025 as an efficiency-focused successor to MiniCPM-V 2.6; support lists llama.cpp, Ollama, vLLM, and SGLang.

4.1B params · dense33K contextApache 2.028.8K downloads465 likes6.3K quant downloads
S

Runs great on a 24 GB GPU

Runs from8 GB Q4_K_M · 32KBest on24 GB Q4_K_M · ~236 tok/s · 32KMacs from8 GB Q4_K_M · ~26 tok/s

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run MiniCPM V 4

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Recommended

Ollama
one command · needs Ollama installed
$ollama run hf.co/openbmb/MiniCPM-V-4-gguf:Q4_K_M

Q4_K_M fits fully on the RTX 4090 · 20.9 GB free · 16K context

Download: ggml-model-Q4_K_M.gguf2.19 GBopenbmb

install, OS, other quants

Ways to runOther ways

each opens step-by-step

Run in cloud

Your own private copy · Q4_K_M · 24 GB GPU

  • The easiest way to try this model
  • Running in a few minutes, easy setup
  • Chat right here or connect any app

$5+ bonus for new users

How Much VRAM Does MiniCPM V 4 Need?

Select a quantization level; the listed hardware adheres to it.

QuantizationVRAM+32KOn RTX 4090 (reference)
2.1 GB2.7 GB2-bit · K-quant — 2-bit quantization with K-quant improvementsFits: 32K context · ~317 tok/s
2.1 GB2.8 GB3-bit small — 3-bit small quantizationFits: 32K context · ~309 tok/s
2.3 GB3 GB3-bit medium — 3-bit medium quantizationFits: 32K context · ~282 tok/s
2.4 GB3 GB4-bit · legacy — 4-bit legacy quantizationFits: 32K context · ~277 tok/s
2.8 GB3.4 GB4-bit medium — 4-bit medium quantization — most popular sweet spotFits: 32K context · ~236 tok/s
3.2 GB3.9 GB5-bit medium — 5-bit medium quantization — good quality/size tradeoffFits: 32K context · ~203 tok/s
3.7 GB4.3 GB6-bit, very good — 6-bit quantization, very good qualityFits: 32K context · ~178 tok/s
4.4 GB5 GB8-bit, near-lossless — 8-bit quantization, near-losslessFits: 32K context · ~149 tok/s
est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

3.1GBat 16K context

Weights
2.4 GB
KV cache
0.3 GB
Overhead
0.3 GB

Fits RTX 4090 (reference) · 20.9 GB free · ~236 tok/s

≥50 tok/s needs ≥214 GB/s · RTX 5090 ~419, M5 Ultra ~309

Which Hardware Can Run MiniCPM V 4?

Q4_K_M · 2.8 GB

MiniCPM V 4 (Q4_K_M) requires 2.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 4+ GB is recommended. Using the full 33K context window can add up to 0.6 GB, bringing total usage to 3.4 GB. 114 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce GTX 1650 SUPER. 185 devices with unified memory can run MiniCPM V 4, including Mac Studio M5 Ultra (256 GB), Mac Studio M5 Ultra (512 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M
8 GB · RTX 4060Q4_K_M · 3.1 GB · ~64 tok/s · 32K8 GB tier →
12 GB · RTX 3060 12GBQ4_K_M · 3.1 GB · ~84 tok/s · 32K12 GB tier →
16 GB · RTX 5060 Ti 16GBQ4_K_M · 3.1 GB · ~105 tok/s · 32K16 GB tier →
24 GB · RTX 4090referenceQ4_K_M · 3.1 GB · ~236 tok/s · 32K24 GB tier →
Mac 32 GB · Mac Mini M4Q4_K_M · 3.1 GB · ~30 tok/s · 32KMacs →
Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 3.1 GB · ~138 tok/s · 32KMacs →

Alternatives to MiniCPM V 4

MiniCPM family · 19 models →

Similar models

6
ModelScoreVRAMtok/sWhy it is similar
Spark X2.6 4BXHToken—4.3 GB~221same size
NVIDIA Nemotron Labs Dory1.0 4B BF16NVIDIA—5.7 GB~214same size
Granite Vision 4.1 4BIBM—4 GB~228same size
Translategemma 4B ITGoogle—5 GB~231same size
Qwen3 4B Instruct 2507Alibaba—4.2 GB~226same size
Phi 4 Mini ReasoningMicrosoft—4.8 GB~228same size

Specifications

Publisher
OpenBMB
Family
MiniCPM
Parameters
4.1B
Architecture
MiniCPMV
Context Length
32,768 tokens
Vocabulary Size
73,448
Release Date
2025-07-12
License
Apache 2.0

Frequently Asked Questions

How much VRAM does MiniCPM V 4 need?

MiniCPM V 4 requires 2.8 GB of VRAM at Q4_K_M, or 8.5 GB at BF16. Full 33K context adds up to 0.6 GB (3.4 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 4.1B × 4.8 bits ÷ 8 = 2.4 GB

KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 1 GB (at full 33K context)

VRAM usage by quantization

2.8 GB
3.4 GB

Learn more about VRAM estimation →

What's the best quantization for MiniCPM V 4?

For MiniCPM V 4, Q4_K_M (2.8 GB) offers the best balance of quality and VRAM usage. Q5_0 (2.9 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 2.1 GB.

VRAM requirement by quantization

Q2_K
2.1 GB
Q3_K_L
2.4 GB
Q4_K_M ★
2.8 GB
Q5_0
2.9 GB
Q5_K_M
3.2 GB
BF16
8.5 GB

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run MiniCPM V 4 on a Mac?

Yes — MacBook Air 13" M3 (8 GB) and 158 other Macs can run MiniCPM V 4. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 2.8 GB of usable unified memory (RAM minus macOS overhead).

Can I run MiniCPM V 4 locally?

Yes — MiniCPM V 4 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is MiniCPM V 4?

At Q4_K_M, MiniCPM V 4 can reach ~1727 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~236 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 2.78 × 0.65 = ~1871 tok/s

Estimated speed at Q4_K_M (2.8 GB)

~1871 tok/s
~236 tok/s
~1871 tok/s
~1727 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of MiniCPM V 4?

At Q4_K_M, the download is about 2.44 GB. The full-precision BF16 version is 8.12 GB. The smallest option (Q2_K) is 1.73 GB.

Which GPUs can run MiniCPM V 4?

114 consumer GPUs can run MiniCPM V 4 at Q4_K_M. Top options include AMD Radeon RX 5600 XT, AMD Radeon RX 5700 XT, AMD Radeon RX 580 8GB, NVIDIA GeForce GTX 1050 Ti. 107 GPUs have plenty of headroom for comfortable inference.

Which devices can run MiniCPM V 4?

185 devices with unified memory can run MiniCPM V 4 at Q4_K_M (2.8 GB), including AMD Ryzen 7 8845HS (Hawk Point) Laptop, AMD Ryzen AI 7 350 (Krackan Point) Laptop, AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10. Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.