QW

Qwen 3.8 flash next heretic 3.78bpw

QwenGGUF by soyaakinoharaUpdated 2026-09-18
Download on Hugging Face
Parameters177B
Context256K
Layers48
Architectureqwen4exp
Smallest quant83.6 GB
Quick answer

Qwen 3.8 flash next heretic 3.78bpw needs about 84.9 GB of memory at its main file (a 83.6 GB file) with an 8K context window, which is more than a 48 GB GPU holds. Longer context needs more: at 32K tokens the KV cache alone adds about 3.2 GB.

Qwen 3.8 Flash Next Heretic 3.78bpw – a 177B parameter vision‑enabled LLM

This model is a 177 billion‑parameter Qwen4Exp architecture with Mixture‑of‑Experts. It supports vision through a mmproj projector and offers a 262144‑token context window. The GGUF file is 83.55 GB and requires about 84.9 GB of VRAM at 8 k context.

Good at

  • Large context handling
  • Vision input support
  • Mixture‑of‑Experts efficiency

Limits

  • Requires >80 GB VRAM for full model
  • No official license details
  • No text‑only quant options

Best for

  • High‑resolution image captioning
  • Long‑form document analysis
  • Complex reasoning with visual data

Who it’s for. Developers building vision‑enabled LLM services that need a large context window and MoE efficiency. Suitable for research labs with high‑end GPU resources.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
—83.6 GB (2 parts) 84.9 GB
Too big Get file →

For 80 GB VRAM, use the full 83.55 GB GGUF file; memory usage is approximate. If VRAM is limited, consider offloading or using a smaller context length, but no other quantized variants are provided.

What GPU do you need for Qwen 3.8 flash next heretic 3.78bpw?

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: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: no quant fits at an 8K context.
  • 32 GB VRAM: no quant fits at an 8K context.
  • 48 GB VRAM: no quant fits at an 8K context.

Specifications

Architecture from the GGUF header

Architectureqwen4exp
Parameters177B (approximate, mixture of experts)
Layers48
Trained context262,144 tokens
KV cache (fp16)96 KB per token

Source

Quantized bysoyaakinohara
FamilyQwen
Licenseother
Downloads (HF)2,128
Last updated2026-09-18

Frequently asked questions about Qwen 3.8 flash next heretic 3.78bpw

How much VRAM does Qwen 3.8 flash next heretic 3.78bpw need?

About 84.9 GB at its main file with an 8K context window. That is the 83.6 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 Qwen 3.8 flash next heretic 3.78bpw on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 84.9 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Qwen 3.8 flash next heretic 3.78bpw GGUF?

It depends on the quant: file is 83.6 GB.

What context length does Qwen 3.8 flash next heretic 3.78bpw 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 3.2 GB at 32K.

Where can I download Qwen 3.8 flash next heretic 3.78bpw GGUF?

The files are on Hugging Face in the soyaakinohara/Qwen-3.8-flash-next-heretic-3.78bpw-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 Qwen 3.8 flash next heretic 3.78bpw use?

Hugging Face lists the license as other. 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 this model on 8 GB VRAM?

No, the model requires approximately 84.9 GB VRAM at 8 k context, far exceeding 8 GB.

Does this model support text only inference?

The model includes a vision projector; text‑only inference is possible but the vision component is optional.

What quantization level is used?

The GGUF file uses a tight mixed‑precision 3.78 bpw quantization derived from BF16.

Other GGUFs of Qwen3.8-Flash-Next-heretic-2

Similar models

Browse all →

More from soyaakinohara

All by soyaakinohara →

More Qwen models

All Qwen →

Other large models

All large models →