What are the hardware requirements for AREX Base?

AREX Base requires approximately 164.2 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed AREX Base and how large is it?

AREX Base is developed by BAAI, a model with 122b 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 AREX Base?

AREX Base is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

AREX Base by BAAI — VRAM 164.2GB

Name: AREX Base
Author: BAAI

AREX Base is an open-source language model by BAAI. Features: 122b LLM, VRAM: 164.2GB, License: apache-2.0, LLM Explorer Score: 0.27.

Arxiv:2607.21461 Agent Base model:finetune:qwen/qwen3... Base model:qwen/qwen3.5-122b-a... Conversational Deep-research En Endpoints compatible Image-text-to-text Long-context Mixture-of-experts Qwen3.5 Qwen3 5 moe Reasoning Region:us Safetensors Sharded Tensorflow Tool-use

Model Card on HF 🤗: https://huggingface.co/BAAI/AREX-Base

AREX Base Parameters and Internals

LLM Name	AREX Base
Repository 🤗	https://huggingface.co/BAAI/AREX-Base
Base Model(s)	Qwen/Qwen3.5-122B-A10B Qwen/Qwen3.5-122B-A10B
Model Size	122b
Required VRAM	164.2 GB
Updated	2026-10-21
Maintainer	BAAI
Model Type	qwen3_5_moe
Model Files	2.6 GB: 1-of-61 5.0 GB: 2-of-61 5.0 GB: 3-of-61 5.0 GB: 4-of-61 3.4 GB: 5-of-61 1.8 GB: 6-of-61 5.0 GB: 7-of-61 5.0 GB: 8-of-61 5.0 GB: 9-of-61 3.4 GB: 10-of-61 1.8 GB: 11-of-61 5.0 GB: 12-of-61 5.0 GB: 13-of-61 5.0 GB: 14-of-61 3.4 GB: 15-of-61 1.8 GB: 16-of-61 5.0 GB: 17-of-61 5.0 GB: 18-of-61 5.0 GB: 19-of-61 3.4 GB: 20-of-61 1.8 GB: 21-of-61 5.0 GB: 22-of-61 5.0 GB: 23-of-61 5.0 GB: 24-of-61 3.4 GB: 25-of-61 1.8 GB: 26-of-61 5.0 GB: 27-of-61 5.0 GB: 28-of-61 5.0 GB: 29-of-61 3.4 GB: 30-of-61 1.8 GB: 31-of-61 5.0 GB: 32-of-61 5.0 GB: 33-of-61 5.0 GB: 34-of-61 3.4 GB: 35-of-61 1.8 GB: 36-of-61 5.0 GB: 37-of-61 5.0 GB: 38-of-61 5.0 GB: 39-of-61 3.4 GB: 40-of-61 1.8 GB: 41-of-61
Supported Languages	en
Model Architecture	Qwen3_5MoeForConditionalGeneration
License	apache-2.0
Transformers Version	4.57.0.dev0

Rank the AREX Base 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 AREX Base

Best Alternatives	Context / RAM	Downloads	Likes
Qwen3.5 122B A10B NVFP4	0K / 83.5 GB	248732	49
Ornith 1.5 122B	0K / 174.2 GB	60210	186
Qwen3.5 122B A10B	0K / 248.6 GB	2139628	605
Ornith 1.5 122B NVFP4	0K / 77.6 GB	36820	20
Qwen3.5 122B A10B	0K / 248.6 GB	1680	8
Ornith 1.5 122B FP8	0K / 90.1 GB	50413	12
Qwen3.5 122B A10B	0K / 248.6 GB	45	1
Qwen3.5 122B A10B Abliterated	0K / 248.6 GB	108	5
Qwen3.5 122B A10B FP8	0K / 126.3 GB	1271983	112
...en3.5 122B A10B Int4 AutoRound	0K / 72 GB	142311	45

Note: green Score (e.g. "73.2") means that the model is better than BAAI/AREX-Base.