Ornith-1.5-9B-GGUF

Ornith-1.5-9B-GGUF

Offers GGUF files of Ornith-1.5-9B, a 9B dense model trained through a self-improvement loop, for single-GPU use. Aimed at developers who want a small model with published coding scores.

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 GGUF files of Ornith-1.5-9B, the smallest model in the Ornith-1.5 family. It is a 9B dense model meant for efficient single-GPU use. Ornith-1.5 builds on Ornith-1.0 by letting the training loop generate its own tasks, build the scaffolds and roll out solutions, then learn through reinforcement learning.

Who it suits: It suits developers who want a compact model for one GPU and who care about coding benchmarks, plus researchers who follow self-improving training methods.

Industry | Who uses it | How they use it | Example
Software | Developers with a single GPU | Run a 9B dense model locally | Try command-line coding tasks like those in Terminal-Bench
Mobile | App teams | Look at the quantized Mobile variant for edge devices | Check whether a phone-sized build fits their app
Research | Teams studying self-improvement | Read how tasks, scaffolds and rollouts are optimized together | Compare Ornith-1.5-9B with Ornith-1.0-9B

People who want a ready chat app can skip it, since the card text gives no run steps beyond the download.

Where to use it:

• A single GPU machine, which the card names as the target
• Edge devices, through the separate quantized Mobile variant the card mentions
• Windows, macOS or Linux, where you can download the files with Python

Use case | Who | What they get back
Local coding help | Developers | A small model with published Terminal-Bench and SWE-bench scores
Benchmark comparison | Evaluators | Scores next to Ornith-1.0-9B, Qwen3.5-9B and larger models
Self-improvement study | Researchers | A model built with a loop that generates its own training tasks

How to set it up: You need these first:

• Python
• The huggingface_hub package
• Disk space for the files, about 4.6 GB at 4-bit as an estimate

1. Install the Hugging Face hub package.
2. Download the model files to your computer.

pip install -U huggingface_hub
hf download ornith-ai/Ornith-1.5-9B-GGUF

How to use it: Download the files and load them in a tool that reads the GGUF format. The card text here has no run commands, so check its model card for loading steps and for the Mobile variant.

Example: A developer could download the files, load them locally and ask for help with a command-line task. The card reports 46.2 for Ornith-1.5-9B on Terminal-Bench 2.1 with Terminus-2.

For newcomers and experts: A GGUF file is a packed copy of a model that is made to run on your own machine. A first small step is to download the files and check the card for how to load them.

Ornith-1.0 was built on Qwen3.5 and Gemma4 with continued pretraining, mid-training and post-training. In Ornith-1.5 the loop also writes new tasks, on top of tuning scaffolds and rollouts. That replaces a fixed set of curated tasks and hand-built frameworks with tasks the model generates. Larger models still score higher on some tests, such as Qwen3.6-35B-A3B on Terminus-2.

New training tasks are generated
Strategies for solving them are found
Solutions are rolled out and rewarded
Reinforcement learning improves the policy

Good to know:

• Size: 9.2B parameters per Hugging Face, about 4.6 GB at 4-bit, as an estimate
• Versions: Hugging Face lists 5 quantized versions made by other people
• Benchmarks: the card reports coding scores against Ornith-1.0-9B, Qwen3.5-9B, Qwen3.6-35B-A3B and Gemma-4-31B

Title: Terminal-Bench 2.1 with Terminus-2
Ornith-1.5-9B: 46.2
Ornith-1.0-9B: 43.1
Qwen3.5-9B: 21.3
Qwen3.6-35B-A3B: 52.5
Gemma-4-31B: 42.1

Title: SWE-bench Verified
Ornith-1.5-9B: 70.6
Ornith-1.0-9B: 69.4
Qwen3.5-9B: 53.2
Qwen3.6-35B-A3B: 73.4

• License: mit
• Cost to run: none named, no API key or account
• Data it reaches: it downloads model files from Hugging Face, and the card text names no other service or account

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 9.2B
• Weights: about 4.6 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 5,772,425
• Likes: 673
• License: mit
• Quantized versions: 5
• Tags: transformers, gguf, text-generation, conversational

Links

• Model: https://huggingface.co/ornith-ai/Ornith-1.5-9B-GGUF
• Files: https://huggingface.co/ornith-ai/Ornith-1.5-9B-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download ornith-ai/Ornith-1.5-9B-GGUF

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