Qwen3.5-27B
Qwen3.5-27B by Alibaba Qwen (27B parameters). At Q4_K_M with an 8K context it needs about 17.0 GiB of memory. A dense 27B with image and video input; its hybrid attention keeps a KV cache in only 16 of 64 layers. It fits a 20–24 GB card at 5-bit.
Last updated: 2026-10-13
Model facts
Run it
ollama run qwen3.5:27b
Memory needed
Weights + KV cache + runtime overhead, in GiB. Lower quantization = less memory, slightly lower quality.
| Quantization | 4K context | 8K context | 32K context |
|---|---|---|---|
| Q4_K_M | 16.8 GiB | 17.0 GiB | 18.5 GiB |
| Q5_K_M | 19.5 GiB | 19.8 GiB | 21.3 GiB |
| Q6_K | 22.5 GiB | 22.7 GiB | 24.2 GiB |
| Q8_0 | 28.8 GiB | 29.1 GiB | 30.6 GiB |
| FP16 | 53.6 GiB | 53.8 GiB | 55.3 GiB |
Which hardware runs it
Best quantization that fits each device with a 8K context. Estimates, not benchmarks. See the methodology.
Frequently asked questions
How much VRAM does Qwen3.5-27B need?
With an 8K context, about 17.0 GiB at Q4_K_M and 29.1 GiB at Q8_0 (weights + KV cache + runtime overhead). Longer contexts need more memory for the KV cache.
Can Qwen3.5-27B run on an GeForce RTX 4090?
Yes. Qwen3.5-27B fits on the GeForce RTX 4090 (24 GB) at Q6_K with an 8K context, using about 22.7 GiB, at an estimated 32 tokens/s.
Can I run Qwen3.5-27B on a Mac?
Yes. Qwen3.5-27B fits on the M3 Ultra (96 GB) at FP16 with an 8K context, using about 53.8 GiB, at an estimated 9.9 tokens/s.
Can Qwen3.5-27B be used commercially?
Qwen3.5-27B is released under Apache-2.0. The licence allows commercial use (read its conditions).
Sources
- https://github.com/QwenLM/Qwen3.8/blob/main/README.md
- https://huggingface.co/Qwen/Qwen3.5-27B/blob/main/LICENSE
- https://huggingface.co/Qwen/Qwen3.5-27B
- https://huggingface.co/Qwen/Qwen3.5-27B/resolve/main/config.json
- https://huggingface.co/api/models/Qwen/Qwen3.5-27B?expand[]=safetensors
- https://ollama.com/library/qwen3.5/tags