Qwen2.5 VL 72B Instruct — Hardware Requirements & GPU Compatibility
VisionQwen2.5-VL-72B-Instruct is Alibaba's 73-billion-parameter instruction-tuned vision-language model and represents the largest offering in the Qwen2.5-VL lineup. The card notes it can recognize text, charts, and layouts within images; act as a visual agent for controlling computers or phones; process videos exceeding an hour while pinpointing events; perform bounding-box and point localization; and generate structured outputs from scanned invoices and tables. Models of this magnitude require multi-GPU setups or machines with ample unified memory, even when quantized. The documentation states the shipped configuration supports 32,768 tokens but describes extending it via YaRN for longer inputs. Released under the Qwen license instead of Apache 2.0, you should review its terms before any commercial application. Published in January 2025, this model succeeds Qwen2-VL and reportedly improves upon that predecessor based on developer feedback.
Doesn't fit a 24 GB GPU
Smallest consumer fit → Snapdragon X Elite Copilot+ PC · IQ2_XXS · 25.9 GB · ~4 tok/s
Macs from 32 GB (IQ2_XS · ~12 tok/s)
Other
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Qwen2.5 VL 72B Instruct
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/ggml-org/Qwen2.5-VL-72B-Instruct-GGUF:Q4_K_MQ4_K_M · needs ~49.7 GB · 16K context
Download: Qwen2.5-VL-72B-Instruct-Q4_K_M.gguf47.42 GBggml-org
Ways to runOther ways
each opens step-by-stepRun in cloud
Your own private copy · Q4_K_M · 96 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
Runs on A100 80GB (30 GB headroom) · Q4_K_M
- Generation speed
- ~29 tok/s
- generation speed
- Cost per 1M output tokens
- $10.73
- per 1M output tokens
How Much VRAM Does Qwen2.5 VL 72B Instruct Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 32.2 GB | 42.2 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Does not fit: needs 36.9 GB | |
| 33.1 GB | 43.2 GB | 3-bit small — 3-bit small quantization | Does not fit: needs 37.8 GB | |
| 36.8 GB | 46.8 GB | 3-bit medium — 3-bit medium quantization | Does not fit: needs 41.5 GB | |
| 37.7 GB | 47.8 GB | 4-bit · legacy — 4-bit legacy quantization | Does not fit: needs 42.4 GB | |
| 45.0 GB | 55.1 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Does not fit: needs 49.7 GB | |
| 53.3 GB | 63.4 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Does not fit: needs 58 GB | |
| 61.5 GB | 71.6 GB | 6-bit, very good — 6-bit quantization, very good quality | Does not fit: needs 66.2 GB | |
| 74.4 GB | 84.5 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Does not fit: needs 79.1 GB |
Selected Q4_K_M
4-bit medium quantization49.7GBat 16K context
- Weights
- 44.1 GB
- KV cache
- 5.4 GB
- Overhead
- 0.3 GB
Doesn't fit RTX 4090 (reference) · needs 25.7 GB more
Or .
Which Hardware Can Run Qwen2.5 VL 72B Instruct?
Q4_K_M · 45.0 GBQwen2.5 VL 72B Instruct (Q4_K_M) requires 45.0 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 59+ GB is recommended. Using the full 128K context window can add up to 41.3 GB, bringing total usage to 86.3 GB. No consumer GPU has enough memory. 56 devices with unified memory can run Qwen2.5 VL 72B Instruct, including Mac Studio M5 Ultra (256 GB), Mac Studio M5 Ultra (512 GB), Mac Pro M2 Ultra (64 GB).
32–40 GB
Q4_K_M · best fit IQ2_M (30.5 GB)
1 GPU · 3 Mac classes · 2 device families
RTX 5090~45Simulate output on RTX 5090M5 Max*36 GB~13Simulate output on Mac Studio M5 Max (36 GB)M632 GB~5Simulate output on iMac 24" M6 (32 GB)Orin~5Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~3Simulate output on Snapdragon X Elite Copilot+ PC64–80 GB
fits · 14.3 GB free · up to 32K ctx
3 Mac classes · 1 device family
M2 Ultra*64 GB~12Simulate output on Mac Pro M2 Ultra (64 GB)M5 Max*64 GB~10Simulate output on Mac Studio M5 Max (64 GB)Orin~3Simulate output on NVIDIA Jetson AGX Orin 64GB
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to Qwen2.5 VL 72B Instruct
Qwen 2.5 family · 45 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Llama 3.3 70B InstructMeta | 79.9 | 51.3 GB | —needs > 24 GB | same size |
| NVIDIA Nemotron 3.5 Bolt 74B A9B BF16NVIDIA | — | 54.2 GB | —needs > 24 GB | same size |
| Kolibri 1 BF16Aleph-Alpha | — | 22.5 GB | ~237 | same size |
| CodeLlama 70B Instruct HFMeta | — | 23.2 GB | ~29 | same size |
| StableBeluga2Stability AI | — | 23.5 GB | ~29 | same size |
| Apertus 70B 2509swiss-ai | — | 48 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- Alibaba
- Family
- Qwen 2.5
- Parameters
- 73.4B
- Architecture
- Qwen2_5_VLForConditionalGeneration
- Context Length
- 128,000 tokens
- Vocabulary Size
- 152,064
- Release Date
- 2025-01-27
- License
- Other
Frequently Asked Questions
How much VRAM does Qwen2.5 VL 72B Instruct need?
Qwen2.5 VL 72B Instruct requires 45.0 GB of VRAM at Q4_K_M, or 147.8 GB at BF16. Full 128K context adds up to 41.3 GB (86.3 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 73.4B × 4.8 bits ÷ 8 = 44 GB
KV Cache + Overhead ≈ 1 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 ≈ 42.3 GB (at full 128K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Qwen2.5 VL 72B Instruct?
Yes, at IQ2_S (23.9 GB) or lower. Higher quantizations like IQ2_M (25.8 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Qwen2.5 VL 72B Instruct?
For Qwen2.5 VL 72B Instruct, Q4_K_M (45.0 GB) offers the best balance of quality and VRAM usage. Q4_K_L (45.9 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 21.2 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Qwen2.5 VL 72B Instruct on a Mac?
Yes — Mac Studio M1 Max (32 GB) and 90 other Macs can run Qwen2.5 VL 72B Instruct, the smallest at IQ2_XS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 45.0 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (64 GB).
Can I run Qwen2.5 VL 72B Instruct locally?
Yes — Qwen2.5 VL 72B Instruct can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Qwen2.5 VL 72B Instruct?
At Q4_K_M, Qwen2.5 VL 72B Instruct can reach ~107 tok/s on AMD Instinct MI350X. 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 ÷ 45.0 × 0.65 = ~116 tok/s
Estimated speed at Q4_K_M (45.0 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Qwen2.5 VL 72B Instruct?
At Q4_K_M, the download is about 44.05 GB. The full-precision BF16 version is 146.82 GB. The smallest option (IQ2_XXS) is 20.19 GB.
Which GPUs can run Qwen2.5 VL 72B Instruct?
No single consumer GPU has enough VRAM to run Qwen2.5 VL 72B Instruct at Q4_K_M (45.0 GB). Multi-GPU or professional hardware is required.
Which devices can run Qwen2.5 VL 72B Instruct?
56 devices with unified memory can run Qwen2.5 VL 72B Instruct at Q4_K_M (45.0 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (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.