Xeretron Omega 320B — NOTICE
=============================

This file must accompany any distribution of the weights, quantizations,
or derivatives of "Xeretron Omega 320B".

The contents of this NOTICE are informational and do not modify the license
of the base Work.


1) Xeretron derivative
----------------------

  Derivative name:      Xeretron Omega 320B
  Operator / producer:  Xeretron (Xeretron AI Server)
  Modification type:    finetune / post-training and operational packaging
                        (Omega series: brain-of-brains — high-quality
                        architectures, hard bugs, device control, and agent
                        orchestration).
  Modification notice:  This work is a Derivative Work of GLM-5.3-Flash
                        (official: zai-org/GLM-5.3-Flash). Files and weights
                        have been modified relative to the original Work.

  Copyright © 2026 Xeretron. All rights reserved in Xeretron's own
  contributions (finetune, packaging, documentation, and the "Xeretron
  Omega 320B" mark), without prejudice to the rights in the base Work.


2) Base Work (upstream)
-----------------------

  Base model:    GLM-5.3-Flash
  Authors:       Z.AI / Zhipu AI (per model card)
  Official:      https://huggingface.co/zai-org/GLM-5.3-Flash
  License:       MIT License — Copyright (c) 2026 Z.AI Co., Ltd
                 See LICENSE-MIT-zai-glm-5.3-flash.en.txt

  Technical summary (public model card):
  - ~320B / 321B total · ~18B active/token · native multimodal MoE
  - Hybrid sparse + linear attention · Manifold-Constrained Hyper-Connections (mHC)
  - Native context 1,048,576 tokens
  - Multimodal: text, image, video · tool calling · reasoning_effort low/high/max
  - Multimodal pretraining ~30T tokens

  Copyright © 2026 Z.AI Co., Ltd (and respective owners).


3) Trademarks
-------------

  "GLM", "GLM-5.3", "Flash", "Z.ai", "Zhipu" and related marks are property
  of their respective owners.

  "Xeretron", "Xeretron Omega", "CodiiX" and "Xeretron AI Server" are marks
  of Xeretron. This derivative does not imply sponsorship or official
  association with Z.AI / Zhipu.


4) Xeretron distribution policy
-------------------------------

  In compliance with the MIT License of the base Work, Xeretron shares
  Omega 320B for whatever use the user wants on their Xeretron node.
  Xeretron does not charge for the model (weights / finetune): when there
  is a commercial relationship, it covers infrastructure, panel,
  installation, and support.

  MIT allows use, modification, distribution, and commercial use of the
  Software subject to retaining the copyright notice and the MIT permission
  notice. Include this NOTICE and LICENSE-MIT-zai-glm-5.3-flash.en.txt with
  every delivery.


5) Packaging
------------

  Include with every delivery:
    - MIT LICENSE of the base Work + this NOTICE
    - modification notice (section 1)
    - metadata: base_model = zai-org/GLM-5.3-Flash

  Throughput claimed in marketing (e.g. ~18 t/s on 2× RTX 3090) is a
  Xeretron measurement; it is not part of the license and is not a warranty.
  Reference node: 2× RTX 3090 (24 GB each), CPU 24 threads / 12 physical,
  ~92 GB system RAM.


6) Contact
----------

  Site:  https://xeretron.com/
  Page:  https://xeretron.com/en/modelo-xeretron-omega-320b.html
  Base:  https://huggingface.co/zai-org/GLM-5.3-Flash
  Z.ai:  https://z.ai/blog/glm-5.3-flash