MiniCPM V 4.6 Thinking — Hardware Requirements & GPU Compatibility
VisionFunctions
MiniCPM-V 4.6 Thinking is OpenBMB's 1.3-billion-parameter vision-language model, a long chain-of-thought reasoning variant of MiniCPM-V 4.6 built on a SigLIP2-400M vision encoder paired with a small Qwen3.5-0.8B language backbone. It generates an explicit reasoning trace before answering, aimed at multimodal reasoning, math, and OCR-heavy document tasks rather than quick captioning, keeping the same edge-friendly, phone-oriented architecture. Its small size lets it run on a single modest consumer GPU. The model supports a 262,144 token context window. It is released under the Apache 2.0 license, permitting unrestricted commercial and research use, and was published in May 2026. Its distinguishing trait versus base 4.6 is the thinking mode, which trades some latency for better performance on reasoning-heavy visual tasks while reusing the same mixed 4x/16x visual token compression.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 256K)Best on 24 GB (Q4_K_M · ~580 tok/s · 256K)Macs from 8 GB (Q4_K_M · ~63 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run MiniCPM V 4.6 Thinking
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 run hf.co/openbmb/MiniCPM-V-4.6-Thinking-gguf:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 22.5 GB free · 16K context
Download: MiniCPM-V-4_6-Thinking-Q4_K_M.gguf0.53 GBopenbmb
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3060 12GB (10 GB headroom) · Q4_K_M
- Generation speed
- ~207 tok/s
- generation speed
- Cost per 1M output tokens
- $0.080
- per 1M output tokens
How Much VRAM Does MiniCPM V 4.6 Thinking Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 0.9 GB | 1.7 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 256K context · ~728 tok/s | |
| 0.9 GB | 1.7 GB | 3-bit small — 3-bit small quantization | Fits: 256K context · ~712 tok/s | |
| 1.0 GB | 1.7 GB | 3-bit medium — 3-bit medium quantization | Fits: 256K context · ~669 tok/s | |
| 1 GB | 1.8 GB | 4-bit · legacy — 4-bit legacy quantization | Fits: 256K context · ~655 tok/s | |
| 1.1 GB | 1.9 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 256K context · ~580 tok/s | |
| 1.3 GB | 2 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 256K context · ~512 tok/s | |
| 1.4 GB | 2.2 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 256K context · ~461 tok/s | |
| 1.6 GB | 2.4 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 256K context · ~397 tok/s |
Selected Q4_K_M
4-bit medium quantization1.5GBat 16K context
- Weights
- 0.8 GB
- KV cache
- 0.4 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 22.5 GB free · ~580 tok/s
≥50 tok/s needs ≥87 GB/s · RTX 5090 ~1031, RTX 3090 Ti ~580
Which Hardware Can Run MiniCPM V 4.6 Thinking?
Q4_K_M · 1.1 GBMiniCPM V 4.6 Thinking (Q4_K_M) requires 1.1 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 2+ GB is recommended. Using the full 262K context window can add up to 6.4 GB, bringing total usage to 7.5 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run MiniCPM V 4.6 Thinking, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 6.5 GB free · up to 256K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~437Simulate output on RTX 3080RTX 2080 Ti~354Simulate output on RTX 2080 TiRTX 3070 Ti~350Simulate output on RTX 3070 TiM3*8 GB~63Simulate output on iMac 24" M3 (8 GB)Orin Nano~59Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~42Simulate output on iPhone 1720 GB
fits · 18.5 GB free · up to 256K ctx
1 GPU
RX 7900 XT~425Simulate output on RX 7900 XT32–40 GB
fits · 30.5 GB free · up to 256K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~1031Simulate output on RTX 5090M5 Max36 GB~285Simulate output on Mac Studio M5 Max (36 GB)M532 GB~95Simulate output on MacBook Air 13" M5 (32 GB)Orin~118Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~78Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to MiniCPM V 4.6 Thinking
MiniCPM family · 18 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Llama 3.2 1B InstructMeta | — | 2.3 GB | ~799 | same size |
| Inkling Small DSpark PreviewRadixArk | — | 1.5 GB | ~555 | same size |
| Sweep Next Edit 1.5BXenova | — | 1.6 GB | ~551 | same size |
| Ilama 3.2 1Bhmellor | — | 3.3 GB | ~231 | same size |
| T5gemma L L Ul2 ITGoogle | — | 4.3 GB | ~240 | same size |
| Granite 3.0 1B A400m InstructIBM | — | 1.3 GB | ~524 | same size |
Specifications
- Publisher
- OpenBMB
- Family
- MiniCPM
- Parameters
- 1.3B
- Architecture
- MiniCPMV4_6ForConditionalGeneration
- Context Length
- 262,144 tokens
- Vocabulary Size
- 248,094
- Release Date
- 2026-05-08
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does MiniCPM V 4.6 Thinking need?
MiniCPM V 4.6 Thinking requires 1.1 GB of VRAM at Q4_K_M, or 3.0 GB at BF16. Full 262K context adds up to 6.4 GB (7.5 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 1.3B × 4.8 bits ÷ 8 = 0.8 GB
KV Cache + Overhead ≈ 0.3 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 ≈ 6.7 GB (at full 262K context)
VRAM usage by quantization
What's the best quantization for MiniCPM V 4.6 Thinking?
For MiniCPM V 4.6 Thinking, Q4_K_M (1.1 GB) offers the best balance of quality and VRAM usage. Q5_0 (1.2 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 0.7 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run MiniCPM V 4.6 Thinking on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run MiniCPM V 4.6 Thinking. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 1.1 GB of usable unified memory (RAM minus macOS overhead).
Can I run MiniCPM V 4.6 Thinking locally?
Yes — MiniCPM V 4.6 Thinking can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is MiniCPM V 4.6 Thinking?
At Q4_K_M, MiniCPM V 4.6 Thinking can reach ~4248 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~580 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 ÷ 1.13 × 0.65 = ~4602 tok/s
Estimated speed at Q4_K_M (1.1 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of MiniCPM V 4.6 Thinking?
At Q4_K_M, the download is about 0.78 GB. The full-precision BF16 version is 2.60 GB. The smallest option (IQ2_XXS) is 0.36 GB.
Which GPUs can run MiniCPM V 4.6 Thinking?
52 consumer GPUs can run MiniCPM V 4.6 Thinking at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.
Which devices can run MiniCPM V 4.6 Thinking?
163 devices with unified memory can run MiniCPM V 4.6 Thinking at Q4_K_M (1.1 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.