What are the hardware requirements for Mira 1 XL?

Mira 1 XL requires approximately 55.4 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Mira 1 XL and how large is it?

Mira 1 XL is developed by Smilyai-labs, a model with 27b 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 Mira 1 XL?

Mira 1 XL is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Mira 1 XL by Smilyai-labs — VRAM 55.4GB

Name: Mira 1 XL
Author: Smilyai-labs

Mira 1 XL is an open-source language model by Smilyai-labs. Features: 27b LLM, VRAM: 55.4GB, License: apache-2.0, LLM Explorer Score: 0.25.

Base model:jackrong/qwen3.5-27... Base model:merge:jackrong/qwen... Base model:merge:qwen/qwen3.5-... Base model:merge:teichai/qwen3... Base model:merge:teichai/qwen3... Base model:qwen/qwen3.5-27b Base model:teichai/qwen3.5-27b... Base model:teichai/qwen3.8-27b... Bf16 Conversational Creative-writing Dare-ties En Endpoints compatible Image-text-to-text Merge Multimodal Qwen3 5 Reasoning Region:us Safetensors Sharded Tensorflow Vision Zh

Model Card on HF 🤗: https://huggingface.co/Smilyai-labs/Mira-1-XL

Mira 1 XL Parameters and Internals

LLM Name	Mira 1 XL
Repository 🤗	https://huggingface.co/Smilyai-labs/Mira-1-XL
Base Model(s)	Qwen/Qwen3.5-27B Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill TeichAI/Qwen3.8-27B-Fable-Distill Qwen/Qwen3.5-27B Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill TeichAI/Qwen3.8-27B-Fable-Distill
Model Size	27b
Required VRAM	55.4 GB
Updated	2026-09-22
Maintainer	Smilyai-labs
Model Type	qwen3_5
Model Files	5.3 GB: 1-of-11 5.3 GB: 2-of-11 5.3 GB: 3-of-11 5.3 GB: 4-of-11 5.3 GB: 5-of-11 5.3 GB: 6-of-11 5.3 GB: 7-of-11 5.3 GB: 8-of-11 5.3 GB: 9-of-11 5.3 GB: 10-of-11 2.4 GB: 11-of-11
Supported Languages	en zh
Model Architecture	Qwen3_5ForConditionalGeneration
License	apache-2.0
Model Max Length	262144
Transformers Version	5.8.0.dev0
Tokenizer Class	Qwen2Tokenizer
Padding Token	<\|endoftext\|>
Errors	replace

Rank the Mira 1 XL 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 Mira 1 XL

Best Alternatives	Context / RAM	Downloads	Likes
...wopus3.8 27B Flash Mlx OQ4e 1M	1024K / 16.9 GB	211	1
Agens Pilot	1024K / 54.7 GB	95	3
Qwen3.8 27B NPU2	256K / GB	252	4
JEV 27B VL	0K / 55.6 GB	1525286	966
Clef	0K / 54.7 GB	20160	2194
Qwen3.6 27B	0K / 55.4 GB	6728138	2197
Qwen3.8 27B	0K / 55.6 GB	2	8753
Qwen3.8 27B	0K / 55.6 GB	2727	21
Qwen3.8 27B MLX	0K / 16 GB	1411	18
Qwen3.6 27B NVFP4	0K / 22 GB	1887525	419

Note: green Score (e.g. "73.2") means that the model is better than Smilyai-labs/Mira-1-XL.