SAVRN
Search Contact SAVRN

Open-weight model · Text generation

Qwen3.5-122B-A10B-NVFP4

by NVIDIA nvidia/Qwen3.5-122B-A10B-NVFP4

The NVIDIA Qwen3.5-122B-A10B-NVFP4 model is the quantized version of Alibaba's Qwen3.5-122B-A10B model, which is an auto-regressive language model that uses an optimized transformer architecture. For more information, please check here.

Parameters64.6B
Context262,144
Weights83.5 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads1.4M

Runs On

What it takes to serve Qwen3.5-122B-A10B-NVFP4 (64.6B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

PrecisionWeightsMemory neededCheapest setupPer hourAlso fits
16-bit 129.2 GB 155.0 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x MI325X $2.00 · 1x MI355X $2.59
8-bit 64.6 GB 77.5 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00
4-bit 32.3 GB 38.7 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026.

SAVRN's Notes on Qwen3.5-122B-A10B-NVFP4

Eight of 256 experts fire per token, so only a slice of the 64.6B parameters is in play at once. At 4-bit the weights are 32.3 GB and the working set 38.7 GB; 8-bit needs 77.5 GB; even 16-bit at 155.0 GB fits the cheapest setup on our Index, one MI300X with 192 GB at $1.85 an hour on-demand. One card covers every precision listed; the decision is how much leftover headroom goes to the 262,144-token context, since long prompts fill what the weights leave.

Apache 2.0 permits commercial use, modification and redistribution, and NVIDIA states it neither owns nor developed the underlying model. Before committing, read the base card for Qwen/Qwen3.5-122B-A10B, since this is quantized from and derived from it, and note that files ship as safetensors only. No host prices are on our Index yet, so your own hardware sets the token cost.

SAVRN Research, 2026-09-18

Model Card

By NVIDIA, published under apache-2.0, revision 98915d837c4e.

Model Overview

Description:

The NVIDIA Qwen3.5-122B-A10B-NVFP4 model is the quantized version of Alibaba's Qwen3.5-122B-A10B model, which is an auto-regressive language model that uses an optimized transformer architecture. For more information, please check here. The NVIDIA Qwen3.5-122B-A10B NVFP4 model is quantized with Model Optimizer.

This model is ready for commercial/non-commercial use.

Third-Party Community Consideration

This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party’s requirements for this application and use case; see link to Non-NVIDIA (Qwen3.5-122B-A10B) Model Card from Alibaba.

References

Nvidia Model Optimizer: https://github.com/NVIDIA/Model-Optimizer

License/Terms of Use:

Apache license 2.0

Deployment Geography:

Global

Use Case:

Developers looking to take off-the-shelf, pre-quantized models for deployment in AI Agent systems, chatbots, RAG systems, and other AI-powered applications.

Release Date:

Huggingface 06/01/2026 via https://huggingface.co/nvidia/Qwen3.5-122B-A10B-NVFP4

Model Architecture:

Architecture Type: Transformers
Network Architecture: Qwen3.5-122B-A10B Mixture of Experts
Number of Model Parameters: 122B in total and 10B activated

Input:

Read the full model card (926 words)

Configuration

Architecture
Qwen3_5MoeForConditionalGeneration
Context length (tokens)
262,144
Layers
48
Hidden size
3,072
Attention heads
32
Key/value heads
2
Head dimension
256
Vocabulary size
248,320
Experts
256
Experts active per token
8
Model type
qwen3_5_moe
Quantization
modelopt

Identity and Version

Repository
nvidia/Qwen3.5-122B-A10B-NVFP4
Publisher
NVIDIA
Task
Text generation
Modality
Text
Library
Model Optimizer
Parameters
64.6B parameters
Languages
Not stated by the source
Revision
98915d837c4e7c87ac8296d02e89de19b3207e6d
First published
2026-05-13
Last updated
2026-06-02

Files and Weights

21 files, 83.5 GB in total. The weights are 9 files totalling 83.5 GB in safetensors.

Weights9 files · 83.5 GB
Configuration6 files · 16.5 MB
Tokenizer3 files · 19.5 MB
Documentation1 file · 8.2 KB
Other1 file · 7.8 KB
Repository1 file · 1.6 KB
Every file
FileTypeSizeSHA-256
model-00001-of-00009.safetensorsWeights10.0 GB 1250c8e43ec0
model-00002-of-00009.safetensorsWeights10.0 GB e82c2c017ade
model-00003-of-00009.safetensorsWeights10.0 GB 9739b910c9cd
model-00004-of-00009.safetensorsWeights10.0 GB 1c971546e7b8
model-00005-of-00009.safetensorsWeights10.0 GB abbf91a68fc4
model-00006-of-00009.safetensorsWeights10.0 GB f3b73a17118b
model-00007-of-00009.safetensorsWeights10.0 GB 1080813452cb
model-00008-of-00009.safetensorsWeights10.0 GB b1b5056e7c0f
model-00009-of-00009.safetensorsWeights3.5 GB ea8a02e3dbb9
config.jsonConfiguration14.3 KB —
generation_config.jsonConfiguration244 B —
hf_quant_config.jsonConfiguration9.6 KB —
model.safetensors.index.jsonConfiguration16.5 MB b0f1f40086d4
preprocessor_config.jsonConfiguration390 B —
video_preprocessor_config.jsonConfiguration385 B —
README.mdDocumentation8.2 KB —
chat_template.jinjaOther7.8 KB —
.gitattributesRepository1.6 KB —
tokenizer.jsonTokenizer12.8 MB 5f9e4d4901a9
tokenizer_config.jsonTokenizer16.7 KB —
vocab.jsonTokenizer6.7 MB —

License and Download

License
apache-2.0
Access
Open weights, no gate
Download size
83.5 GB
Download from NVIDIA

Released by NVIDIA through its official repository on Hugging Face. Read the license.

Built From

  • Derived from Qwen/Qwen3.5-122B-A10B
  • Quantized from Qwen/Qwen3.5-122B-A10B

Memory Requirements

PrecisionWeights in memory
As published83.5 GB
16-bit129.2 GB
8-bit64.6 GB
4-bit32.3 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Questions About Qwen3.5-122B-A10B-NVFP4

How much GPU memory does Qwen3.5-122B-A10B-NVFP4 need?

About 155 GB at 16-bit and 38.7 GB at 4-bit: the weights (64.6B parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run Qwen3.5-122B-A10B-NVFP4 on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use Qwen3.5-122B-A10B-NVFP4 commercially?

Yes. Qwen3.5-122B-A10B-NVFP4 is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

What is Qwen3.5-122B-A10B-NVFP4's context length?

262,144 tokens, from the maximum position embeddings in its published configuration.

Similar Models

For more details on how to deploy and use the model - see the Quick Start Guide below! The post-training data has a cutoff date of February 2026. The pre-training data has a cutoff date of June 2025. NVIDIA Nemotron™ is a family of open models with open weights, training data, and recipes, delivering leading efficiency and accuracy for building specialized AI agents. Nemotron-3-Super-120B-A12B-NVFP4 is a large language model (LLM) trained by NVIDIA, designed to deliver strong agentic, reasoning, and conversational capabilities. It is optimized for collaborative agents and high-volume workloads such as IT ticket automation. Like other models in the family, it responds to user queries and…

Open weights other 67.2B parameters 262,144 tokens transformers
View model

Model · Text generation

Llama-3.3-70B-Instruct

Meta Llama

The Meta Llama 3.3 multilingual large language model (LLM) is an instruction tuned generative model in 70B (text in/text out). The Llama 3.3 instruction tuned text only model is optimized for multilingual dialogue use cases and outperforms many of the available open source and closed chat models on common industry benchmarks. Model Architecture: Llama 3.3 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety. Supported languages: English, German, French, Italian, Portuguese, Hindi, Spanish, and…

Access requested at publisher llama3.3 70.6B parameters transformers
View model

Model · Text generation

Qwen-72B

Qwen

通义千问-72B（Qwen-72B）是阿里云研发的通义千问大模型系列的720亿参数规模的模型。Qwen-72B是基于Transformer的大语言模型, 在超大规模的预训练数据上进行训练得到。预训练数据类型多样，覆盖广泛，包括大量网络文本、专业书籍、代码等。同时，在Qwen-72B的基础上，我们使用对齐机制打造了基于大语言模型的AI助手Qwen-72B-Chat。本仓库为Qwen-72B的仓库。 通义千问-72B（Qwen-72B）主要有以下特点： 1. 大规模高质量训练语料：使用超过3万亿tokens的数据进行预训练，包含高质量中、英、多语言、代码、数学等数据，涵盖通用及专业领域的训练语料。通过大量对比实验对预训练语料分布进行了优化。 2. 强大的性能：Qwen-72B在多个中英文下游评测任务上（涵盖常识推理、代码、数学、翻译等），效果显著超越现有的开源模型。具体评测结果请详见下文。 3. 覆盖更全面的词表：相比目前以中英词表为主的开源模型，Qwen-72B使用了约15万大小的词表。该词表对多语言更加友好，方便用户在不扩展词表的情况下对部分语种进行能力增强和扩展。 4. 较长的上下文支持：Qwen-72B支持32k的上下文长度。 Qwen-72B is the 72B-parameter version of the large language model series, Qwen (abbr. Tongyi Qianwen), proposed by Alibaba Cloud. Qwen-72B is a Transformer-based large…

Open weights other 72.3B parameters 32,768 tokens transformers
View model

Model · Text generation

Qwen3-Coder-Next-FP8

Qwen

Today, we're announcing Qwen3-Coder-Next-FP8, an open-weight language model designed specifically for coding agents and local development. It features the following key enhancements: Qwen3-Coder-Next-FP8 has the following features: NOTE: This model supports only non-thinking mode and does not generate blocks in its output. Meanwhile, specifying enablethinking=False is no longer required. For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our blog, GitHub, and Documentation. We advise you to use the latest version of transformers. The following contains a code snippet illustrating how to use the model generate content based on…

Open weights apache-2.0 79.7B parameters 262,144 tokens transformers
View model

Model · Text generation

Qwen3.6-35B-A3B-abliterated-v4

CS

Uncensored version of Qwen/Qwen3.6-35B-A3B with refusal behavior removed via abliteration (norm-preserving orthogonalization). Zero refusals on harmful prompts. No false refusals on harmless prompts. Abliteration identifies the "refusal direction" in the model's residual stream — the linear direction that activates when the model decides to refuse — and surgically removes it from all output projection weights using norm-preserving orthogonalization. 1. Collect residual stream activations (last token position) for 512 harmful + 512 harmless prompts across all 40 layers 2. Compute mean difference vector per layer → this is the "refusal direction" candidate 3. Score layers by…

Open weights apache-2.0 34.7B parameters 262,144 tokens transformers
View model

Model · Text generation

dolphin-2.9.1-yi-1.5-34b

Dolphin

Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations This is our most spectacular outcome ever. FFT, all parameters, 16bit. 77.4 MMLU on 34b. And it talks like a dream. Although the max positional embeddings is 4k, we used rope theta of 1000000.0 and we trained with sequence length 8k. We plan to train on the upcoming 32k version as well. Our appreciation for the sponsors of Dolphin 2.9.1: - Crusoe Cloud - provided excellent on-demand 8xH100 node - OnDemand - provided inference sponsorship This model is based on Yi-1.5-34b, and is governed by apache 2.0 license. The base model has 4k context, but we used rope theta of 1000000.0 and the…

Open weights apache-2.0 34.4B parameters 8,192 tokens transformers
View model