Ornith-1.5-122B-NVFP4

Ornith-1.5-122B-NVFP4

Gives you an NVFP4 quantized copy of Ornith-1.5-122B, a 122B mixture-of-experts text model meant for single-GPU use and strong in coding tests. For developers who host large models themselves.

Made by ornith-ai on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: This repository holds a quantized NVFP4 copy of Ornith-1.5-122B, a text generation model from ornith-ai. The card describes the original as a 122B mixture-of-experts model designed to run on a single GPU, with a separate mobile variant for phones.

Ornith-1.5 builds on Ornith-1.0, which started from Qwen3.5 and Gemma4. Its training loop makes new tasks, builds the scaffold to solve them and learns from the attempts through reinforcement learning.

Who it suits: It suits developers and teams who self-host a large model for coding and agent work and want a smaller quantized copy to fit on fewer GPUs.

Industry | Who uses it | How they use it | Example
Software | Developers | Host a coding model on their own server | Fix a failing build from a terminal
AI platforms | ML engineers | Serve a quantized large model to internal users | Run a private coding assistant
Research | Model researchers | Study self-improving training loops | Compare against other large models
Security | Security engineers | Try the model on cyber benchmarks | Run the card's CyberGym style tests

If you do not have a large GPU machine, or you need a model for images or audio, you can skip it.

Where to use it: You fetch the files on Windows, macOS or Linux with Python. A 64 GB model needs a machine with plenty of GPU memory to run.

• A server or workstation with large GPU memory
• The transformers library on the Hugging Face side
• A self-hosted coding or agent setup

Use case | Who | What they get back
Coding help | Developers | Code and fixes written as text
Agent tasks | Builders | Step-by-step text output for tools and terminals
Model testing | Researchers | Scores to compare with other models

How to set it up: You need Python with pip to fetch the files, and about 64 GB of disk space for the weights. The card does not give run commands in the part shown here, so follow its own instructions to run it.

• Python with pip
• About 64 GB of free disk space
• A machine with large GPU memory

1. Install the Hugging Face hub package.
2. Download the repository files.
3. Load the model with a runtime that reads NVFP4 weights.

pip install -U huggingface_hub
hf download ornith-ai/Ornith-1.5-122B-NVFP4

How to use it: Once the model is loaded, you send text prompts and read the text it writes. Coding and terminal tasks are where the card points its results.

Example: You give the model a failing command and the error text and ask for a fix. It answers with a written explanation and the changed code or command.

For newcomers and experts: If you are new, know that this is a huge download meant for strong hardware. Start by reading the model card, then check that your machine has the memory for 64 GB of weights.

If you know the field, the training loop tunes task generation, scaffold construction and solution rollouts together. It improves the policy with reinforcement learning and does not rely on a fixed set of human-written tasks.

Generate new training tasks
Build a scaffold to solve them
Run solution rollouts
Improve the policy with RL

Good to know: The card compares the 122B model with several other models, and some score higher on some tests.

Benchmark | Ornith-1.5-122B | Qwen3.8-Flash-Next | Qwen3.8 27B
TerminalBench 2.1 | 81.2 | 83.4 | 71.9
CyberGym | 80 | 78 | 55

• License: mit
• Cost to run: none named, but it needs large GPU memory
• Data it reaches: files you download, and the prompts you send

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 64B
• Weights: about 64 GB in U8
• Formats: Safetensors
• Downloads in the last 30 days: 279,663
• Likes: 36
• License: mit
• Quantized versions: 2
• Tags: transformers, safetensors, qwen3_5_moe, image-text-to-text, text-generation, conversational, 8-bit, modelopt

Links

• Model: https://huggingface.co/ornith-ai/Ornith-1.5-122B-NVFP4
• Files: https://huggingface.co/ornith-ai/Ornith-1.5-122B-NVFP4/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download ornith-ai/Ornith-1.5-122B-NVFP4

Then follow the usage in its model card. Nothing in this kit runs by itself.