LLM EXPLORER 63,407 MODELS INDEXED

AREX 2 27B INT8 W8A16 MTP by numsu

By numsu · 592 downloads
HF Repo ↗ Copy URL

AREX 2 27B INT8 W8A16 MTP is an open-source language model by numsu. Features: 27b LLM, VRAM: 30.7GB, License: apache-2.0, LLM Explorer Score: 0.31.

  Arxiv:2609.38288   Agent   Arex-2   Base model:baai/arex-2 Base model:quantized:baai/arex...   Compressed-tensors   Conversational   Image-text-to-text   Long-context   Mtp   Qwen3 5   Reasoning   Region:us   Safetensors   Sharded   Tensorflow   Tool-use   Vllm   W8a16
Model Card on HF 🤗: https://huggingface.co/numsu/AREX-2-27B-INT8-W8A16-MTP 

AREX 2 27B INT8 W8A16 MTP Parameters and Internals

LLM NameAREX 2 27B INT8 W8A16 MTP
Repository 🤗https://huggingface.co/numsu/AREX-2-27B-INT8-W8A16-MTP 
Base Model(s)  AREX 2   BAAI/AREX-2
Model Size27b
Required VRAM30.7 GB
Updated2026-10-08
Maintainernumsu
Model Typeqwen3_5
Model Files  4.4 GB: 1-of-11   2.7 GB: 2-of-11   2.7 GB: 3-of-11   2.7 GB: 4-of-11   2.7 GB: 5-of-11   2.7 GB: 6-of-11   2.7 GB: 7-of-11   2.7 GB: 8-of-11   2.7 GB: 9-of-11   2.2 GB: 10-of-11   2.5 GB: 11-of-11   1.3 GB   0.0 GB
Model ArchitectureQwen3_5ForConditionalGeneration
Licenseapache-2.0
Model Max Length262144
Transformers Version5.2.0
Tokenizer ClassTokenizersBackend
Padding Token<|endoftext|>
Errorsreplace

Best Alternatives to AREX 2 27B INT8 W8A16 MTP

Best Alternatives
Context / RAM
Downloads
Likes
...wopus3.8 27B Flash Mlx OQ4e 1M1024K / 16.9 GB2111
Agens Pilot1024K / 54.7 GB873
Qwen3.8 27B NPU2256K /  GB2524
JEV 27B VL0K / 55.6 GB1525286966
Clef0K / 54.7 GB95131835
Qwen3.6 27B0K / 55.4 GB67281382197
Qwen3.8 27B0K / 55.6 GB28753
Qwen3.8 27B0K / 55.6 GB272721
Qwen3.8 27B MLX0K / 16 GB141118
Qwen3.8 27B FP80K / 0.4 GB123157416
Note: green Score (e.g. "73.2") means that the model is better than numsu/AREX-2-27B-INT8-W8A16-MTP.