Qwen3.8 Flash Next needs about 4.1 GB of memory at Q4_K_M (a 2.8 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 3 quants, from 2.8 GB (Q4_K_M) to 93.5 GB. Longer context needs more: at 32K tokens the KV cache alone adds about 3.3 GB.
Qwen3.8 Flash Next – 51.2 B parameter model with vision and MTP speculative decoding
Qwen3.8 Flash Next is a 51.2 B parameter model in the Qwen family, built on the qwen4exp architecture. It supports vision, uses a 262144‑token context, and is released under the Apache‑2.0 license.
Good at
- Vision inference
- Long‑context generation
- MTP speculative decoding
Limits
- Requires a custom llama.cpp fork
- No host‑RAM fallback for n‑gram table
- Large VRAM needed for full model
Best for
- Vision‑enabled local inference
- Agentic loops with long prompts
- High‑throughput speculative decoding
Who it’s for. Developers and researchers who need a local, vision‑capable LLM with long‑context support and speculative decoding. The model runs entirely on GPU and requires a Vulkan‑capable GPU with sufficient VRAM.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 2.8 GB | 4.1 GB | Fits | Get file → | |
| — | 22.4 GB | 23.7 GB | Too big | Get file → | |
| — | 93.5 GB | 94.8 GB | Too big | Get file → |
For 8 GB VRAM use the Q4_K_M quantized file (2.79 GB, ~4.1 GB memory at 8 k context). For larger VRAM budgets the 22.4 GB or 93.48 GB files provide higher‑precision options, but memory usage scales accordingly.
What GPU do you need for Qwen3.8 Flash Next?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: the best quant that fits is Q4_K_M (2.8 GB file, about 4.1 GB in use).
- 8 GB VRAM: the best quant that fits is Q4_K_M (2.8 GB file, about 4.1 GB in use).
- 12 GB VRAM: the best quant that fits is Q4_K_M (2.8 GB file, about 4.1 GB in use).
- 16 GB VRAM: the best quant that fits is Q4_K_M (2.8 GB file, about 4.1 GB in use).
- 24 GB VRAM: the best quant that fits is the main file (22.4 GB file, about 23.7 GB in use).
- 32 GB VRAM: the best quant that fits is the main file (22.4 GB file, about 23.7 GB in use).
- 48 GB VRAM: the best quant that fits is the main file (22.4 GB file, about 23.7 GB in use).
Which Qwen3.8 Flash Next quant should you choose?
Q4_K_M is the usual starting point for Qwen3.8 Flash Next: it balances file size and quality well. Here it is a 2.8 GB file.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Frequently asked questions about Qwen3.8 Flash Next
How much VRAM does Qwen3.8 Flash Next need?
About 4.1 GB at Q4_K_M with an 8K context window. That is the 2.8 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Qwen3.8 Flash Next on an 8 GB GPU?
Yes, with Q4_K_M: the 2.8 GB file needs about 4.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Qwen3.8 Flash Next GGUF?
It depends on the quant: Q4_K_M is 2.8 GB, file is 22.4 GB, file is 93.5 GB.
What context length does Qwen3.8 Flash Next support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.8 GB at 8K and 3.3 GB at 32K.
Where can I download Qwen3.8 Flash Next GGUF?
The files are on Hugging Face in the TonyMFJ/Qwen3.8-Flash-Next-WinVulkan-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Qwen3.8 Flash Next use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Qwen3.8 Flash Next on 8 GB VRAM?
Yes, using the Q4_K_M quantized file (2.79 GB) which requires about 4.1 GB of VRAM at 8 k context.
Does this model support vision?
Yes, it includes a vision tower at f16, losslessly converted.
Do I need a special llama.cpp fork?
Yes, the model requires the agentionai/llama.cpp fork for ROCmFPx quant types and per‑head PLE layout.