What are the hardware requirements for Qwen3.5 397B A17B NVFP4 V2?

Qwen3.5 397B A17B NVFP4 V2 requires approximately 243.5 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Qwen3.5 397B A17B NVFP4 V2 and how large is it?

Qwen3.5 397B A17B NVFP4 V2 is developed by nvidia, a model with 397b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Qwen3.5 397B A17B NVFP4 V2 perform on standard benchmarks?

Qwen3.5 397B A17B NVFP4 V2 has the following published scores: LMSYS ELO 1442. Compare against reference models on this page or on the LLM Explorer leaderboards.

Qwen3.5 397B A17B NVFP4 V2 by nvidia — VRAM 243.5GB

Name: Qwen3.5 397B A17B NVFP4 V2
Author: nvidia

Qwen3.5 397B A17B NVFP4 V2 is an open-source language model by nvidia. Features: 397b LLM, VRAM: 243.5GB, License: apache-2.0, LLM Explorer Score: 0.47, ELO: 1442.

Base model:quantized:qwen/qwen... Base model:qwen/qwen3.5-397b-a... Conversational Fp4 Model optimizer Modelopt Modelopt mixed Nvidia Quantized Qwen3.5 Qwen3 5 moe Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/nvidia/Qwen3.5-397B-A17B-NVFP4-V2

Qwen3.5 397B A17B NVFP4 V2 Benchmarks

LMSys ELO 1442 vs 1272 (so35) 13.4%

^nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

What is the LLM Explorer Rank (Score)

Qwen3.5 397B A17B NVFP4 V2 Parameters and Internals

LLM Name	Qwen3.5 397B A17B NVFP4 V2
Repository 🤗	https://huggingface.co/nvidia/Qwen3.5-397B-A17B-NVFP4-V2
Base Model(s)	Qwen/Qwen3.5-397B-A17B Qwen/Qwen3.5-397B-A17B
Model Size	397b
Required VRAM	243.5 GB
Updated	2026-07-12
Maintainer	nvidia
Model Type	qwen3_5_moe
Model Files	10.0 GB: 1-of-26 10.0 GB: 2-of-26 10.0 GB: 3-of-26 10.0 GB: 4-of-26 10.0 GB: 5-of-26 10.0 GB: 6-of-26 10.0 GB: 7-of-26 10.0 GB: 8-of-26 10.0 GB: 9-of-26 10.0 GB: 10-of-26 10.0 GB: 11-of-26 10.0 GB: 12-of-26 10.0 GB: 13-of-26 10.0 GB: 14-of-26 10.0 GB: 15-of-26 10.0 GB: 16-of-26 10.0 GB: 17-of-26 10.0 GB: 18-of-26 10.0 GB: 19-of-26 10.0 GB: 20-of-26 10.0 GB: 21-of-26 10.0 GB: 22-of-26 8.3 GB: 23-of-26 2.0 GB: 24-of-26 10.0 GB: 25-of-26 3.2 GB: 26-of-26
Model Architecture	Qwen3_5MoeForConditionalGeneration
License	apache-2.0
Model Max Length	262144
Transformers Version	5.10.2
Tokenizer Class	Qwen2Tokenizer
Padding Token	<\|im_end\|>
Errors	replace

Rank the Qwen3.5 397B A17B NVFP4 V2 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 Qwen3.5 397B A17B NVFP4 V2

Best Alternatives	Context / RAM	Downloads	Likes
Qwen3.5 397B A17B	0K / 369.8 GB	331990	1549
Qwen3.5 397B A17B NVFP4	0K / 251.1 GB	428565	102
Ornith 1.0 397B	0K / 288.5 GB	318396	260
Ornith 1.0 397B FP8	0K / 147.7 GB	597119	183
XYZ Aquila Pro	0K / 225 GB	1805	375
Qwen3.5 397B A17B	0K / 369.8 GB	53	13
Qwen3.5 397B A17B FP8	0K / 184.9 GB	825515	182
Qwen3.5 397B A17B MXFP4	0K / 98.9 GB	21989	6
Ornith 1.0 397B	0K / 288.5 GB	447	4
...7B A17B Heretic Int4 AutoRound	0K / 212.3 GB	4351	4

Note: green Score (e.g. "73.2") means that the model is better than nvidia/Qwen3.5-397B-A17B-NVFP4-V2.