SAVRN
Search Contact SAVRN

Open-weight model · Image and text to text

Aura-Prototype-26B-A4B

by EldritchLabs EldritchLabs/Aura-Prototype-26B-A4B

This is a merge of pre-trained language models created using mergekit. This model was merged using the aura merge method. Aura is an experimental method with a live heatmap visualizer.

Parameters26B
Context262,144
Weights52.0 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads—

Runs On

What it takes to serve Aura-Prototype-26B-A4B (26B 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 52.0 GB 62.4 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00
8-bit 26.0 GB 31.2 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00
4-bit 13.0 GB 15.6 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.

Model Card

By EldritchLabs, published under apache-2.0, revision 264a1400d00b.

This is a merge of pre-trained language models created using mergekit. This model was merged using the aura merge method. Aura is an experimental method with a live heatmap visualizer. This model took 10 hours to merge using graphv18.py The following models were included in the merge: - TheDrummer/Orion-26B-A4B-v1.1 - Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2 - electroglyph/gemma4-26b-fiction-bf16 The following YAML configuration was used to produce this model

Read EldritchLabs's full model card

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the aura merge method.

Aura is an experimental method with a live heatmap visualizer. This model took 10 hours to merge using graph_v18.py

Models Merged

The following models were included in the merge: - Gandalf69/Adversary-26B-A4B-v0 - TheDrummer/Orion-26B-A4B-v1.1 - Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2 - electroglyph/gemma4-26b-fiction-bf16

Configuration

The following YAML configuration was used to produce this model:

architecture: Gemma4ForConditionalGeneration
models:
- model: B:\26B\electroglyph--gemma4-26b-fiction-bf16
parameters:
weight: 1.0
density: 0.90
epsilon: 0.09
pinocchio: false
- model: B:\26B\Gryphe--Pantheon-Reasoning-26B-A4B-1.1-V2
parameters:
weight: 1.0
density: 0.90
epsilon: 0.09
pinocchio: false
- model: B:\26B\Gandalf69--Adversary-26B-A4B-v0
parameters:
weight: 1.0
density: 0.90
epsilon: 0.09
pinocchio: true
- model: B:\26B\TheDrummer--Orion-26B-A4B-v1.1
parameters:
weight: 1.0
density: 0.90
epsilon: 0.09
pinocchio: false
merge_method: aura
parameters:
eta: 1.2
tol: 1.0e-9
max_iter: 32
kappa: 0.65
mu: 0.5
theta: 0.94
optimization_mode: "pareto"
phi: 2.0
alpha: 0.4
kappa: 0.65
jl_projs: 4
halley_strength: 0.20
halley_period: 76
scalar_gravity: 0.05
aura_shift: true
omega: 0.08
momentum: 0.25
patience: 64
sharp: 0.55
voice: 0.35
density: 0.90
epsilon: 0.09
rescale: true
int8_mask: false
normalize_weights: false
normalize_router: true
router_strategy: "della"
blend_experts: true
auto_pad: true
skip_on_mismatch: true
verbose_logging: true
resume_path: "B:/mergekit-new/Aura_Cache"
enable_qliphoth: true
pinocchio_lambda: 0.65
inverse_pinocchio: false
dynamic_pareto_pinocchio: true
qliphoth_gamma: 0.40
qliphoth_lambda_enslaved: 0.15
qliphoth_svd_rank: 4
qliphoth_preserve_ratio: 0.70
save_layer_heatmaps: true
audit_image_dir: "B:/mergekit-new/Aura_Heatmaps"
gap_jumper: 0.75
enable_qhe: true
qhe_levels: 16
qhe_intensity: 0.15
qhe_crossover: 0.15
dtype: float32
out_dtype: bfloat16
tokenizer:
source: union
chat_template: auto
name: Aura Prototype 26B-A4B

Configuration

Architecture
Gemma4ForConditionalGeneration
Context length (tokens)
262,144
Layers
30
Hidden size
2,816
Feed-forward size
2,112
Attention heads
16
Key/value heads
8
Head dimension
256
Vocabulary size
262,144
Experts
128
Sliding window (tokens)
1,024
Model type
gemma4

Identity and Version

Repository
EldritchLabs/Aura-Prototype-26B-A4B
Publisher
EldritchLabs
Task
Image and text to text
Modality
Image and text
Library
transformers
Parameters
26B parameters
Languages
eng
Revision
264a1400d00b187e5540a07457790dbcb7e19a5b
First published
2026-09-18
Last updated
2026-09-18

Files and Weights

23 files, 52.0 GB in total. The weights are 11 files totalling 52.0 GB in safetensors.

Weights11 files · 52.0 GB
Configuration6 files · 74.5 KB
Tokenizer2 files · 32.2 MB
Documentation1 file · 3.6 KB
Other2 files · 578.9 KB
Repository1 file · 1.6 KB
Every file
FileTypeSizeSHA-256
model-00001-of-00011.safetensorsWeights4.6 GB d3b64c8523dd
model-00002-of-00011.safetensorsWeights4.9 GB 5ffb59418767
model-00003-of-00011.safetensorsWeights4.9 GB 8684e4c6eb82
model-00004-of-00011.safetensorsWeights4.9 GB 96fe8de5d741
model-00005-of-00011.safetensorsWeights4.9 GB a962fc5c6ff6
model-00006-of-00011.safetensorsWeights4.9 GB 80822b341095
model-00007-of-00011.safetensorsWeights4.9 GB 637d4e3782ca
model-00008-of-00011.safetensorsWeights4.9 GB bca987f07759
model-00009-of-00011.safetensorsWeights4.9 GB 8bab1fcb846a
model-00010-of-00011.safetensorsWeights4.9 GB b29e8c3ec613
model-00011-of-00011.safetensorsWeights3.3 GB ab3d794499bc
aura_progress.jsonConfiguration48 B —
config.jsonConfiguration4.0 KB —
crucible_override.jsonConfiguration79 B —
crucible_telemetry.jsonConfiguration2.7 KB —
mergekit_config.ymlConfiguration1.9 KB —
model.safetensors.index.jsonConfiguration65.8 KB —
README.mdDocumentation3.6 KB —
AuraMerge.pngOther559.8 KB a09a9e5b8ba7
chat_template.jinjaOther19.1 KB —
.gitattributesRepository1.6 KB —
tokenizer.jsonTokenizer32.2 MB cc8d3a0ce364
tokenizer_config.jsonTokenizer3.9 KB —

License and Download

License
apache-2.0
Access
Open weights, no gate
Download size
52.0 GB
Download from EldritchLabs

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

Built From

  • Derived from Gandalf69/Adversary-26B-A4B-v0
  • Derived from Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2
  • Derived from TheDrummer/Orion-26B-A4B-v1.1
  • Derived from electroglyph/gemma4-26b-fiction-bf16
  • Merged from Gandalf69/Adversary-26B-A4B-v0
  • Merged from Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2
  • Merged from TheDrummer/Orion-26B-A4B-v1.1
  • Merged from electroglyph/gemma4-26b-fiction-bf16

Memory Requirements

PrecisionWeights in memory
As published52.0 GB
16-bit52.0 GB
8-bit26.0 GB
4-bit13.0 GB

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

Questions About Aura-Prototype-26B-A4B

How much GPU memory does Aura-Prototype-26B-A4B need?

About 62.4 GB at 16-bit and 15.6 GB at 4-bit: the weights (26B parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run Aura-Prototype-26B-A4B 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 Aura-Prototype-26B-A4B commercially?

Yes. Aura-Prototype-26B-A4B 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 Aura-Prototype-26B-A4B's context length?

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

Similar Models

Model · Image and text to text

gemma-4-26B-A4B-it

Google

Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages. Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in five distinct sizes: E2B, E4B, 12B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from…

Open weights apache-2.0 25.8B parameters 262,144 tokens transformers
View model

Model · Image and text to text

gemma-4-26B-A4B-it-AWQ-4bit

Cyankiwi

Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on small models) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages. Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in four distinct sizes: E2B, E4B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from high-end phones…

Open weights apache-2.0 25.8B parameters 262,144 tokens transformers
View model

Model · Image and text to text

Qwen3.8-27B-MLX-4bit

LM Studio Community

LM Studio Community models highlights program. Highlighting new & noteworthy models by the community. Join the conversation on Discord. 4-bit quantized version of Qwen3.8-27B using MLX, optimized for Apple Silicon. Special thanks to the Apple Machine Learning Research team for creating MLX. LM Studio is not the creator, originator, or owner of any Model featured in the Community Model Program. Each Community Model is created and provided by third parties. LM Studio does not endorse, support, represent or guarantee the completeness, truthfulness, accuracy, or reliability of any Community Model. You understand that Community Models can produce content that might be offensive, harmful…

Open weights apache-2.0 27.4B parameters 262,144 tokens transformers
View model

Model · Image and text to text

Qwen3.8-27B-MLX-8bit

LM Studio Community

LM Studio Community models highlights program. Highlighting new & noteworthy models by the community. Join the conversation on Discord. 8-bit quantized version of Qwen3.8-27B using MLX, optimized for Apple Silicon. Special thanks to the Apple Machine Learning Research team for creating MLX. LM Studio is not the creator, originator, or owner of any Model featured in the Community Model Program. Each Community Model is created and provided by third parties. LM Studio does not endorse, support, represent or guarantee the completeness, truthfulness, accuracy, or reliability of any Community Model. You understand that Community Models can produce content that might be offensive, harmful…

Open weights apache-2.0 27.4B parameters 262,144 tokens transformers
View model

Model · Image and text to text

Qwen3.8-27B-MLX-6bit

LM Studio Community

LM Studio Community models highlights program. Highlighting new & noteworthy models by the community. Join the conversation on Discord. 6-bit quantized version of Qwen3.8-27B using MLX, optimized for Apple Silicon. Special thanks to the Apple Machine Learning Research team for creating MLX. LM Studio is not the creator, originator, or owner of any Model featured in the Community Model Program. Each Community Model is created and provided by third parties. LM Studio does not endorse, support, represent or guarantee the completeness, truthfulness, accuracy, or reliability of any Community Model. You understand that Community Models can produce content that might be offensive, harmful…

Open weights apache-2.0 27.4B parameters 262,144 tokens transformers
View model