llmrun
NVIDIAQwen 3.5Qwen3_5ForCausalLM

Qwen3.5 4B Qwen3.8 27B Draft — Hardware Requirements & GPU Compatibility

ChatFunctions

Qwen3.5 4B Qwen3.8 27B Draft is a 4.2B-parameter open language model from NVIDIA in the Qwen 3.5 family. It supports a context window of up to 262,144 tokens. At Q4_K_M it needs about 2.99 GB of VRAM — see which GPUs and Macs can run it below.

4.2B params · dense262K contextOther1.1K downloads232 likes
Based on Qwen3.5 4B
S

Runs great on a 24 GB GPU

Runs from8 GB Q4_K_M · 128KBest on24 GB Q4_K_M · ~219 tok/s · 256KMacs from8 GB Q4_K_M · ~24 tok/s

Other

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Qwen3.5 4B Qwen3.8 27B Draft

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

vLLM
one command · needs vLLM installed
$vllm serve nvidia/Qwen3.5-4B-Qwen3.8-27B-Draft --max-model-len 16384

vLLM repo with its own quantisation

Download: Weights on Hugging Face

install, OS, other quants

Ways to runOther ways

each opens step-by-step

Run in cloud

Runs on RTX 3060 12GB (8 GB headroom) · Q4_K_M

Generation speed
~78 tok/s
generation speed
Cost per 1M output tokens
$0.21
per 1M output tokens

How Much VRAM Does Qwen3.5 4B Qwen3.8 27B Draft Need?

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

QuantizationVRAM+32KOn RTX 4090 (reference)
2.3 GB2.8 GB2-bit · K-quant — 2-bit quantization with K-quant improvementsFits: 256K context · ~290 tok/s
2.5 GB3 GB3-bit medium — 3-bit medium quantizationFits: 256K context · ~260 tok/s
3.0 GB3.5 GB4-bit medium — 4-bit medium quantization — most popular sweet spotFits: 256K context · ~219 tok/s
3.5 GB4 GB5-bit medium — 5-bit medium quantization — good quality/size tradeoffFits: 256K context · ~189 tok/s
3.9 GB4.4 GB6-bit, very good — 6-bit quantization, very good qualityFits: 256K context · ~166 tok/s
4.7 GB5.2 GB8-bit, near-lossless — 8-bit quantization, near-losslessFits: 256K context · ~140 tok/s
8.9 GB9.4 GB16-bit · bfloat — Brain floating point 16 — preferred for trainingFits: 256K context · ~74 tok/s
est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

3.2GBat 16K context

Weights
2.5 GB
KV cache
0.3 GB
Overhead
0.3 GB

Fits RTX 4090 (reference) · 20.8 GB free · ~219 tok/s

≥50 tok/s needs ≥230 GB/s · RTX 5090 ~390, M5 Ultra ~288

Which Hardware Can Run Qwen3.5 4B Qwen3.8 27B Draft?

Q4_K_M · 3.0 GB

Qwen3.5 4B Qwen3.8 27B Draft (Q4_K_M) requires 3.0 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 262K context window can add up to 5.2 GB, bringing total usage to 8.2 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 Qwen3.5 4B Qwen3.8 27B Draft, 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 · 3.2 GB · ~59 tok/s · 128K8 GB tier →
12 GB · RTX 3060 12GBQ4_K_M · 3.2 GB · ~78 tok/s · 256K12 GB tier →
16 GB · RTX 5060 Ti 16GBQ4_K_M · 3.2 GB · ~97 tok/s · 256K16 GB tier →
24 GB · RTX 4090referenceQ4_K_M · 3.2 GB · ~219 tok/s · 256K24 GB tier →
Mac 32 GB · Mac Mini M4Q4_K_M · 3.2 GB · ~28 tok/s · 256KMacs →
Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 3.2 GB · ~128 tok/s · 256KMacs →

Alternatives to Qwen3.5 4B Qwen3.8 27B Draft

Qwen 3.5 family · 35 models →

Similar models

6
ModelScoreVRAMtok/sWhy it is similar
Gemma 3 4B ITGoogle64.05 GB~231same size
Spark X2.6 4BXHToken—4.3 GB~221same size
Granite Vision 4.1 4BIBM—4 GB~228same size
Translategemma 4B ITGoogle—5 GB~231same size
Qwen3 VL 4B InstructAlibaba—4.5 GB~208same size
MiniCPM V 4OpenBMB—3.1 GB~236same size

Specifications

Publisher
NVIDIA
Family
Qwen 3.5
Parameters
4.2B
Architecture
Qwen3_5ForCausalLM
Context Length
262,144 tokens
Vocabulary Size
248,320
Release Date
2026-11-11
License
Other

Frequently Asked Questions

How much VRAM does Qwen3.5 4B Qwen3.8 27B Draft need?

Qwen3.5 4B Qwen3.8 27B Draft requires 3.0 GB of VRAM at Q4_K_M, or 8.9 GB at BF16. Full 262K context adds up to 5.2 GB (8.2 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 4.2B × 4.8 bits ÷ 8 = 2.5 GB

KV Cache + Overhead ≈ 0.5 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 ≈ 5.7 GB (at full 262K context)

VRAM usage by quantization

3.0 GB
8.2 GB

Learn more about VRAM estimation →

What's the best quantization for Qwen3.5 4B Qwen3.8 27B Draft?

For Qwen3.5 4B Qwen3.8 27B Draft, Q4_K_M (3.0 GB) offers the best balance of quality and VRAM usage. Q5_K_M (3.5 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 2.3 GB.

VRAM requirement by quantization

Q2_K
2.3 GB
Q4_K_M ★
3.0 GB
Q5_K_M
3.5 GB
Q6_K
3.9 GB
Q8_0
4.7 GB
BF16
8.9 GB

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Qwen3.5 4B Qwen3.8 27B Draft on a Mac?

Yes — MacBook Air 13" M3 (8 GB) and 158 other Macs can run Qwen3.5 4B Qwen3.8 27B Draft. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 3.0 GB of usable unified memory (RAM minus macOS overhead).

Can I run Qwen3.5 4B Qwen3.8 27B Draft locally?

Yes — Qwen3.5 4B Qwen3.8 27B Draft can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Qwen3.5 4B Qwen3.8 27B Draft?

At Q4_K_M, Qwen3.5 4B Qwen3.8 27B Draft can reach ~1605 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~219 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 ÷ 2.99 × 0.65 = ~1739 tok/s

Estimated speed at Q4_K_M (3.0 GB)

~1739 tok/s
~219 tok/s
~1739 tok/s
~1605 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 Qwen3.5 4B Qwen3.8 27B Draft?

At Q4_K_M, the download is about 2.52 GB. The full-precision BF16 version is 8.41 GB. The smallest option (Q2_K) is 1.79 GB.

Which GPUs can run Qwen3.5 4B Qwen3.8 27B Draft?

114 consumer GPUs can run Qwen3.5 4B Qwen3.8 27B Draft 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 Qwen3.5 4B Qwen3.8 27B Draft?

185 devices with unified memory can run Qwen3.5 4B Qwen3.8 27B Draft at Q4_K_M (3.0 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.