What are the hardware requirements for Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE?

Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE requires approximately 62 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE and how large is it?

Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE is developed by densohax, a model with 72.8b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE?

Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE by densohax — VRAM 62GB

Name: Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE
Author: densohax

Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE is an open-source language model by densohax. Features: 72.8b LLM, VRAM: 62GB, License: other, LLM Explorer Score: 0.34.

256k-context 4-bit Auto-round Autoround Base model:quantized:qwen/qwen... Base model:qwen/qwen3.8-flash-... Conversational Cpu-offload Dual-gpu Fp8 Int4 Local-llm Qwen3.8 Qwen4 exp Region:us Rtx-3090 Safetensors Sharded Speculative-decoding Tensorflow Vllm W4a16

Model Card on HF 🤗: https://huggingface.co/densohax/Qwen3.8-Flash-Next-heretic2-W4A16-FP8PLE

Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE Parameters and Internals

LLM Name	Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE
Repository 🤗	https://huggingface.co/densohax/Qwen3.8-Flash-Next-heretic2-W4A16-FP8PLE
Base Model(s)	Qwen/Qwen3.8-Flash-Next trohrbaugh/Qwen3.8-Flash-Next-heretic-2 RadixArk/Qwen3.8-Flash-Next-NVFP4 Qwen/Qwen3.8-Flash-Next trohrbaugh/Qwen3.8-Flash-Next-heretic-2 RadixArk/Qwen3.8-Flash-Next-NVFP4
Model Size	72.8b
Required VRAM	62 GB
Updated	2026-09-24
Maintainer	densohax
Model Type	qwen4_exp
Model Files	1.5 GB: 1-of-49 1.5 GB: 2-of-49 1.5 GB: 3-of-49 1.4 GB: 4-of-49 1.5 GB: 5-of-49 1.5 GB: 6-of-49 1.5 GB: 7-of-49 1.4 GB: 8-of-49 1.5 GB: 9-of-49 1.5 GB: 10-of-49 1.5 GB: 11-of-49 1.4 GB: 12-of-49 1.5 GB: 13-of-49 1.5 GB: 14-of-49 1.5 GB: 15-of-49 1.4 GB: 16-of-49 1.5 GB: 17-of-49 1.5 GB: 18-of-49 1.5 GB: 19-of-49 1.4 GB: 20-of-49 1.5 GB: 21-of-49 1.5 GB: 22-of-49 1.5 GB: 23-of-49 1.4 GB: 24-of-49 1.5 GB: 25-of-49 1.5 GB: 26-of-49 1.5 GB: 27-of-49 1.4 GB: 28-of-49 1.5 GB: 29-of-49 1.5 GB: 30-of-49 1.5 GB: 31-of-49 1.4 GB: 32-of-49 1.5 GB: 33-of-49 1.5 GB: 34-of-49 1.5 GB: 35-of-49 1.4 GB: 36-of-49 1.5 GB: 37-of-49 1.5 GB: 38-of-49 1.5 GB: 39-of-49 1.4 GB: 40-of-49 1.5 GB: 41-of-49 1.5 GB: 42-of-49
Model Architecture	Qwen4ExpForConditionalGeneration
License	other
Transformers Version	5.17.0

Rank the Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to Qwen3.8 Flash Next Heretic2 W4A16 FP8PLE

Best Alternatives	Context / RAM	Downloads	Likes
...wen3.8 Flash Next W4A16 FP8PLE	0K / 73.7 GB	51	2