AlibabaQwen 3.8Qwen4ExpForConditionalGeneration

Qwen3.8 Flash Next — Hardware Requirements & GPU Compatibility

VisionFunctions

Qwen3.8 Flash Next is a vision-capable model in Alibaba's Qwen 3.8 line, with roughly 180 billion parameters. The Flash Next naming points to a fast-inference variant within the family, and the model accepts image input alongside text for combined visual and textual tasks, extending the Qwen line into multimodal territory. It provides a 256K token context window for long documents or extended sessions, and is released under a custom license outside standard open-source terms. At around 180 billion parameters, running Qwen3.8 Flash Next locally needs multi-GPU or server-class hardware rather than a single consumer GPU, even for quantized deployments.

180.0B params · MoE62B active262K contextOtherAug 20262.4M downloads6.3K likes2.2M quant downloads

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

Qwen3.8 Flash Next runs on a 128 GB+ Mac or a datacenter GPU

Best local pick → Mac Pro M2 Ultra (128 GB) · Q4_K_M · 109 GB · ~13 tok/s

Macs from 64 GB (IQ2_S · ~23 tok/s)

Other

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Qwen3.8 Flash Next

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Other ways

each opens step-by-step

Run in cloud

Fits on H200 (141 GB) (32 GB headroom) · Q4_K_M

Generation speed: ~69 tok/s; generation speed
Cost per 1M output tokens: $14.47; per 1M output tokens

Deployon RunPodBest optionfrom $3.59/hr

Compare GPUs →

Run Qwen3.8 Flash Next

109.0 GB VRAM at 16K context · doesn't fit the RTX 4090 (85.0 GB short)

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF@Q4_K_M
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:Q4_K_M -c 16384 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf unsloth/Qwen3.8-Flash-Next-GGUF:Q4_K_M -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve Qwen/Qwen3.8-Flash-Next --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~396 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model TensorFold/Qwen3.8-Flash-Next-MLX-4bit
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model TensorFold/Qwen3.8-Flash-Next-MLX-4bit

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$unsloth/Qwen3.8-Flash-Next-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
H200 SXM (141 GB)CheapestCheapest32 GB free	~69 tok/s	$14.47	Deploy on RunPodfrom $3.59/hr (opens in new tab)	141 GB · 32 GB free · ~69 tok/s · $14.47 / 1M tokens
H200 NVL (141 GB)32 GB free	~69 tok/s	$16.73	Deploy on Vast.aifrom $4.15/hr (opens in new tab)	141 GB · 32 GB free · ~69 tok/s · $16.73 / 1M tokens
H200 SXM (141 GB)32 GB free	~69 tok/s	$19.00	Deploy on Vast.aifrom $4.71/hr (opens in new tab)	141 GB · 32 GB free · ~69 tok/s · $19.00 / 1M tokens
B200 (192 GB)83 GB free	~102 tok/s	$16.22	Deploy on RunPodfrom $5.98/hr (opens in new tab)	192 GB · 83 GB free · ~102 tok/s · $16.22 / 1M tokens
B300 (288 GB)179 GB free	~102 tok/s	$18.83	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 179 GB free · ~102 tok/s · $18.83 / 1M tokens

Compare GPUs →

How Much VRAM Does Qwen3.8 Flash Next Need?

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

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
49.9 GB	51.1 GB	55.2 GB	~49.5 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Does not fit: needs 50.5 GB
54.4 GB	55.6 GB	59.7 GB	77.7 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Does not fit: needs 55 GB
56.6 GB	57.9 GB	61.9 GB	77.8 GB	2-bit · i-quant small — Importance-weighted 2-bit, small	Does not fit: needs 57.2 GB
61.1 GB	62.4 GB	66.4 GB	80.4 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Does not fit: needs 61.7 GB
70.1 GB	71.4 GB	75.4 GB	82 GB	3-bit · i-quant — Importance-weighted 3-bit	Does not fit: needs 70.7 GB
74.6 GB	75.9 GB	79.9 GB	91.9 GB	3-bit · i-quant extra small — Importance-weighted 3-bit, extra small	Does not fit: needs 75.2 GB
76.9 GB	78.1 GB	82.2 GB	88.2 GB	3-bit · i-quant small — Importance-weighted 3-bit, small	Does not fit: needs 77.5 GB
76.9 GB	78.1 GB	82.2 GB	80.9 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Does not fit: needs 77.5 GB
79.1 GB	80.4 GB	84.4 GB	89.4 GB	3-bit small — 3-bit small quantization	Does not fit: needs 79.7 GB
81.4 GB	82.6 GB	86.7 GB	93.2 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Does not fit: needs 82 GB
88.1 GB	89.4 GB	93.4 GB	92 GB	3-bit medium — 3-bit medium quantization	Does not fit: needs 88.7 GB
90.4 GB	91.6 GB	95.7 GB	100.6 GB	4-bit · legacy — 4-bit legacy quantization	Does not fit: needs 91 GB
92.6 GB	93.9 GB	97.9 GB	93.2 GB	3-bit large — 3-bit large quantization	Does not fit: needs 93.2 GB
97.1 GB	98.4 GB	102.4 GB	93.7 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Does not fit: needs 97.7 GB
101.6 GB	102.9 GB	106.9 GB	111.3 GB	4-bit · legacy+ — 4-bit legacy quantization with offset	Does not fit: needs 102.2 GB
101.6 GB	102.9 GB	106.9 GB	113.1 GB	4-bit small — 4-bit small quantization	Does not fit: needs 102.2 GB
101.6 GB	102.9 GB	106.9 GB	100.3 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Does not fit: needs 102.2 GB
108.4 GB	109.6 GB	113.7 GB	~108 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Does not fit: needs 109 GB
110.6 GB	111.9 GB	115.9 GB	139.3 GB	4-bit large — 4-bit large quantization	Does not fit: needs 111.2 GB
124.1 GB	125.4 GB	129.4 GB	128.1 GB	5-bit small — 5-bit small quantization	Does not fit: needs 124.7 GB
128.6 GB	129.9 GB	133.9 GB	110.5 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Does not fit: needs 129.2 GB
130.9 GB	132.1 GB	136.2 GB	151.1 GB	5-bit large — 5-bit large quantization	Does not fit: needs 131.5 GB
148.9 GB	150.1 GB	154.2 GB	168.1 GB	6-bit, very good — 6-bit quantization, very good quality	Does not fit: needs 149.5 GB
180.4 GB	181.6 GB	185.7 GB	~180 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Does not fit: needs 181 GB
360.4 GB	361.6 GB	365.7 GB	~360 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Does not fit: needs 361 GB

Selected Q4_K_M

4-bit medium quantization

109.0GBat 16K context

Weights: 108.0 GB
KV cache: 0.7 GB
Overhead: 0.3 GB

Doesn't fit RTX 4090 (reference) · needs 85.0 GB more

≥10 tok/s needs ≥583 GB/s · M3 Ultra ~14, M2 Ultra ~13

Or rent H200 from $3.59/hr.

Which Hardware Can Run Qwen3.8 Flash Next?

Q4_K_M · 108.4 GB

Qwen3.8 Flash Next (Q4_K_M) requires 108.4 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 141+ GB is recommended. Using the full 262K context window can add up to 10.7 GB, bringing total usage to 119.0 GB. No consumer GPU has enough memory. 18 devices with unified memory can run Qwen3.8 Flash Next, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (128 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8–16 GB · RTX 4060, RTX 3060 12GB, RTX 5060 Ti 16GBDoesn't fit — smallest quant needs 50.5 GB8 GB tier →12 GB tier →16 GB tier →

24 GB · RTX 4090referenceDoesn't fit — smallest quant needs 50.5 GB24 GB tier →

Mac 32 GB · Mac Mini M4Doesn't fit — smallest quant needs 50.5 GBMacs →

Mac 64 GB · Mac Studio M4 MaxIQ2_S · 57.2 GB · ~17 tok/s · 64K57.2~1764KMacs →

How good is Qwen3.8 Flash Next?

llmrun Score blends public benchmark boards into one number; 100 = the average open model.

Not enough data yet for an llmrun Score.

A Score needs results on enough independent boards. How the Composite Score is built

Benchmarks

LiveBench Coding
LiveBench · Jun 2026Independent
72.5
#12/19 open
Independent
LiveBench Reasoning
LiveBench · Jun 2026Independent
87.4
#2/19 open
Independent
LMArena WebDev
LMArena · Oct 2026Arena
1630.3
#4/52 open
Arena
LiveBench Math
LiveBench · Jun 2026Independent
85.8
#12/19 open
Independent

All 4 Qwen3.8 Flash Next benchmarks

Alternatives to Qwen3.8 Flash Next

Qwen 3.8 family · 12 models →

Similar models

1tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
BloomBigScience	—	122.4 GB	—needs > 24 GB	same size

Fine-tunes & variants(14)community builds of Qwen3.8 Flash Next

Specifications

Publisher: Alibaba
Family: Qwen 3.8
Parameters: 180.0B
Active parameters: 61.6B per token
Architecture: Qwen4ExpForConditionalGeneration
Context Length: 262,144 tokens
Vocabulary Size: 248,320
Release Date: 2026-08-24
License: Other

Resources

Frequently Asked Questions

How much VRAM does Qwen3.8 Flash Next need?

Qwen3.8 Flash Next requires 108.4 GB of VRAM at Q4_K_M, or 360.4 GB at BF16. Full 262K context adds up to 10.7 GB (119.0 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 180.0B × 4.8 bits ÷ 8 = 108 GB

KV Cache + Overhead ≈ 0.4 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 ≈ 11 GB (at full 262K context)

VRAM usage by quantization

Q4_K_M

108.4 GB

Q4_K_M + full context

119.0 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run Qwen3.8 Flash Next?

No — Qwen3.8 Flash Next requires at least 49.9 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for Qwen3.8 Flash Next?

For Qwen3.8 Flash Next, Q4_K_M (108.4 GB) offers the best balance of quality and VRAM usage. Q4_K_L (110.6 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 49.9 GB.

VRAM requirement by quantization

IQ2_XXS

49.9 GB~53%

IQ3_S

76.9 GB~75%

Q3_K_L

92.6 GB~86%

Q4_K_M ★

108.4 GB~89%

Q4_K_L

110.6 GB~90%

BF16

360.4 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Qwen3.8 Flash Next on a Mac?

Yes — Mac Pro M2 Ultra (64 GB) and 41 other Macs can run Qwen3.8 Flash Next, the smallest at IQ2_S. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 108.4 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (128 GB).

Can I run Qwen3.8 Flash Next locally?

Yes — Qwen3.8 Flash Next can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Qwen3.8 Flash Next?

At Q4_K_M, Qwen3.8 Flash Next can reach ~58 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)

Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.

Example: NVIDIA B200 → 37.1 GB active ÷ (8000 × 0.65) = 7.13 ms, plus 48 layers × 0.055 ms = 2.64 ms, so 1000 ÷ 9.77 ms = ~102 tok/s

Estimated speed at Q4_K_M (108.4 GB)

NVIDIA B200

~102 tok/s

NVIDIA B300

~102 tok/s

NVIDIA GH200 Grace Hopper Superchip

~70 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 Qwen3.8 Flash Next?

At Q4_K_M, the download is about 108.00 GB. The full-precision BF16 version is 360.00 GB. The smallest option (IQ2_XXS) is 49.50 GB.

Which GPUs can run Qwen3.8 Flash Next?

No single consumer GPU has enough VRAM to run Qwen3.8 Flash Next at Q4_K_M (108.4 GB). Multi-GPU or professional hardware is required.

Which devices can run Qwen3.8 Flash Next?

18 devices with unified memory can run Qwen3.8 Flash Next at Q4_K_M (108.4 GB), including ASUS Ascent GX10, Mac Pro M2 Ultra (128 GB), Mac Pro M2 Ultra (192 GB), Mac Studio (M3 Ultra, 256GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.