What are the hardware requirements for Smaug Flash?

Smaug Flash requires approximately 148.7 GB of VRAM and supports a context window of 1024K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Smaug Flash and how large is it?

Smaug Flash is developed by abacusai, a model with 304.2b 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 Smaug Flash?

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

Smaug Flash by abacusai — VRAM 148.7GB, 1024K context

Name: Smaug Flash
Author: abacusai

Smaug Flash is an open-source language model by abacusai. Features: 304.2b LLM, VRAM: 148.7GB, Context: 1024K, License: mit.

8-bit Abacusai Agentic Base model:deepseek-ai/deepsee... Base model:finetune:deepseek-a... Code Deepseek v4 Endpoints compatible Fp8 Region:us Safetensors Sharded Smaug Tensorflow

Model Card on HF 🤗: https://huggingface.co/abacusai/Smaug-Flash

Smaug Flash Parameters and Internals

LLM Name	Smaug Flash
Repository 🤗	https://huggingface.co/abacusai/Smaug-Flash
Base Model(s)	DeepSeek V4 Flash 0731 deepseek-ai/DeepSeek-V4-Flash-0731
Model Size	304.2b
Required VRAM	148.7 GB
Updated	2026-09-14
Maintainer	abacusai
Model Type	deepseek_v4
Model Files	1.1 GB: 1-of-48 3.6 GB: 2-of-48 3.6 GB: 3-of-48 3.6 GB: 4-of-48 3.6 GB: 5-of-48 3.6 GB: 6-of-48 3.6 GB: 7-of-48 3.6 GB: 8-of-48 3.6 GB: 9-of-48 3.6 GB: 10-of-48 3.6 GB: 11-of-48 3.6 GB: 12-of-48 3.6 GB: 13-of-48 3.6 GB: 14-of-48 3.6 GB: 15-of-48 3.6 GB: 16-of-48 3.6 GB: 17-of-48 3.6 GB: 18-of-48 3.6 GB: 19-of-48 3.6 GB: 20-of-48 3.6 GB: 21-of-48 3.6 GB: 22-of-48 3.6 GB: 23-of-48 3.6 GB: 24-of-48 3.6 GB: 25-of-48 3.6 GB: 26-of-48 3.6 GB: 27-of-48 3.6 GB: 28-of-48 3.6 GB: 29-of-48 3.6 GB: 30-of-48 3.6 GB: 31-of-48 3.6 GB: 32-of-48 3.6 GB: 33-of-48 3.6 GB: 34-of-48 3.6 GB: 35-of-48 3.6 GB: 36-of-48 3.6 GB: 37-of-48 3.6 GB: 38-of-48 3.6 GB: 39-of-48 3.6 GB: 40-of-48 3.6 GB: 41-of-48 3.6 GB: 42-of-48
Model Architecture	DeepseekV4ForCausalLM
License	mit
Context Length	1048576
Model Max Length	1048576
Transformers Version	4.57.1
Vocabulary Size	129280
Torch Data Type	bfloat16

Rank the Smaug Flash 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 Smaug Flash

Best Alternatives	Context / RAM	Downloads	Likes
DeepSeek V4 Flash 0731	1024K / 152.3 GB	785771	2803
... V4 Flash 0731 Abliterated FP8	1024K / 145.1 GB	5149	35
DeepSeek V4 Flash 0731 NVFP4	1024K / 149.3 GB	7421	6
DeepSeek V4 Flash 0731 NVFP4	1024K / 156.9 GB	307	23
...eepSeek V4 Flash 0731 Ornstein	1024K / 568.7 GB	183	1
...4 Flash 0731 Abliterated NVFP4	1024K / 145.5 GB	952	11
...Seek V4 Flash 0731 CRACK NVFP4	1024K / 141.5 GB	242	7
DeepSeek V4 Flash 0731	1024K / 148.7 GB	89	25
DeepSeek V4 Flash 0731 NVFP4	1024K / 145.5 GB	166	0
DeepSeek V4 Flash 0731 NVFP4	1024K / 156.9 GB	75	2

Note: green Score (e.g. "73.2") means that the model is better than abacusai/Smaug-Flash.