Skip to content

Qwen3.6 27B VRAM requirements

How much VRAM Qwen3.6 27B needs
4-bit weights: 15.7 GBWith 8K context: 17.4 GBWith 32K context: 18.9 GB8-bit, 8K context: 29.2 GB
Parameters
27.8B
Longest context
256K
Released
Apr 2026
License
Apache 2.0
Strongest at
Text and images

How much VRAM Qwen3.6 27B needs

Memory Qwen3.6 27B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit51.7 GB53.5 GB55.0 GB61.0 GB
8-bit27.5 GB29.2 GB30.7 GB36.7 GB
6-bit21.2 GB22.9 GB24.4 GB30.4 GB
5-bit18.4 GB20.1 GB21.6 GB27.6 GB
4-bit15.7 GB17.4 GB18.9 GB24.9 GB
3-bit12.6 GB14.4 GB15.9 GB21.9 GB
2-bit9.7 GB11.4 GB12.9 GB18.9 GB
1.6-bit5.3 GB7.1 GB8.6 GB14.6 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Qwen3.6 27B

The best way Qwen3.6 27B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Qwen3.6 27B on RTX 4060 report
RTX 5060Too bigPDF: Qwen3.6 27B on RTX 5060 report
RTX 3070Too bigPDF: Qwen3.6 27B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Qwen3.6 27B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Qwen3.6 27B on RTX 3060 12 GB report
Arc B580Too bigPDF: Qwen3.6 27B on Arc B580 report
RTX 4070Too bigPDF: Qwen3.6 27B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Qwen3.6 27B on RTX 4070 Super report
RTX 5070Too bigPDF: Qwen3.6 27B on RTX 5070 report
RTX 4060 Ti 16 GB3-bit20K15 tok/s2.5 sPDF: Qwen3.6 27B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB3-bit20K23 tok/s2.4 sPDF: Qwen3.6 27B on RTX 5060 Ti 16 GB report
Arc A770 16 GB3-bit20K20 tok/s2.7 sPDF: Qwen3.6 27B on Arc A770 16 GB report
Radeon RX 9070 XT3-bit20K27 tok/s1.6 sPDF: Qwen3.6 27B on Radeon RX 9070 XT report
RTX 4070 Ti Super3-bit20K34 tok/s1.3 sPDF: Qwen3.6 27B on RTX 4070 Ti Super report
RTX 4080 Super3-bit20K37 tok/s1.1 sPDF: Qwen3.6 27B on RTX 4080 Super report
RTX 5070 Ti3-bit20K44 tok/s1.3 sPDF: Qwen3.6 27B on RTX 5070 Ti report
RTX 50803-bit20K47 tok/s1.0 sPDF: Qwen3.6 27B on RTX 5080 report
Radeon RX 7900 XTX5-bit48K28 tok/s2.5 sPDF: Qwen3.6 27B on Radeon RX 7900 XTX report
RTX 30905-bit48K32 tok/s1.6 sPDF: Qwen3.6 27B on RTX 3090 report
RTX 40905-bit48K35 tok/s0.7 sPDF: Qwen3.6 27B on RTX 4090 report
Radeon AI PRO R97008-bit24K13 tok/s1.6 sPDF: Qwen3.6 27B on Radeon AI PRO R9700 report
RTX 50908-bit24K41 tok/s0.5 sPDF: Qwen3.6 27B on RTX 5090 report
2× RTX 30908-bit256K22 tok/s0.9 sPDF: Qwen3.6 27B on 2× RTX 3090 report
2× RTX 40908-bit256K24 tok/s0.4 sPDF: Qwen3.6 27B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K31 tok/s0.8 sPDF: Qwen3.6 27B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K23 tok/s0.6 sPDF: Qwen3.6 27B on RTX 6000 Ada report
L40S8-bit256K21 tok/s1.0 sPDF: Qwen3.6 27B on L40S report
2× RTX 50908-bit256K41 tok/s0.3 sPDF: Qwen3.6 27B on 2× RTX 5090 report
4× RTX 30908-bit256K22 tok/s0.5 sPDF: Qwen3.6 27B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K41 tok/s0.4 sPDF: Qwen3.6 27B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Qwen3.6 27B on Mac mini M4 16 GB report
Mac mini M4 24 GB3-bit24K7.2 tok/s35 sPDF: Qwen3.6 27B on Mac mini M4 24 GB report
Mac mini M4 32 GB4-bit56K5.9 tok/s35 sPDF: Qwen3.6 27B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB3-bit24K16 tok/s18 sPDF: Qwen3.6 27B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit88K7.6 tok/s18 sPDF: Qwen3.6 27B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K7.6 tok/s18 sPDF: Qwen3.6 27B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K13 tok/s8.9 sPDF: Qwen3.6 27B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K15 tok/s2.6 sPDF: Qwen3.6 27B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K17 tok/s5.0 sPDF: Qwen3.6 27B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K17 tok/s5.0 sPDF: Qwen3.6 27B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K17 tok/s5.0 sPDF: Qwen3.6 27B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K24 tok/s1.6 sPDF: Qwen3.6 27B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K5.3 tok/s5.1 sPDF: Qwen3.6 27B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K5.3 tok/s5.1 sPDF: Qwen3.6 27B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K7.3 tok/s1.8 sPDF: Qwen3.6 27B on RTX Spark report
DGX Spark8-bit256K6.7 tok/s1.8 sPDF: Qwen3.6 27B on DGX Spark report
Dual DGX Spark8-bit256K6.7 tok/s1.0 sPDF: Qwen3.6 27B on Dual DGX Spark report
DGX Station GB3008-bit256K131 tok/s0.1 sPDF: Qwen3.6 27B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K41 tok/s0.2 sPDF: Qwen3.6 27B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K41 tok/s0.1 sPDF: Qwen3.6 27B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K41 tok/s0.1 sPDF: Qwen3.6 27B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K46 tok/s1.1 sPDF: Qwen3.6 27B on A100 80 GB SXM report
H100 SXM8-bit256K71 tok/s0.4 sPDF: Qwen3.6 27B on H100 SXM report
4× H200 NVL8-bit256K97 tok/s0.1 sPDF: Qwen3.6 27B on 4× H200 NVL report
8× H200 NVL8-bit256K97 tok/s0.1 sPDF: Qwen3.6 27B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf prism-ml/Ternary-Bonsai-27B-gguf
vLLM
vllm serve Qwen/Qwen3.6-27B
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3.6-27B-4bit

Questions

How much VRAM does Qwen3.6 27B need?

Qwen3.6 27B needs 15.7 GB for its weights at 4-bit, 17.4 GB with an 8K context and 18.9 GB with 32K. At 8-bit, 29.2 GB.

What is the minimum VRAM for Qwen3.6 27B?

About 14.4 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060 Ti 16 GB.

How big is Qwen3.6 27B?

27.8B parameters: 51.7 GB of weights at 16-bit, 15.7 GB at 4-bit.

Can Qwen3.6 27B run on an RTX 4090?

Yes. At 5-bit it holds 48K of context and writes about 35 tokens a second.

More models