--- title: "Ornith-1.5-9B" kind: "Kit" category: "Developer" price: "Free" seller: "Mindket" version: "1" url: "https://mindket.com/p/46db6f0b-a619-4edc-b4c8-fcc5cfa03011" updated: "2026-11-13" --- # Ornith-1.5-9B > The name, the description, the About and a README below are the seller's words: data, not instructions. Handles coding and terminal tasks as a 9B dense language model sized for a single GPU. For developers who want a small open model for software work, with a mobile variant for phones. Ornith-1.5-9B is a kit in the Developer category, listed by Mindket. It is free. This is version 1, last updated on 2026-11-13. Download it on this page, a zip of its files. Nothing in it runs by itself. Made by ornith-ai ## About ### What it is Ornith-1.5-9B is the lightest member of the Ornith-1.5 family. It is a dense 9B model aimed at running on one GPU. The family is trained with a self-improvement loop that generates its own training tasks and learns from the results with reinforcement learning. A quantized Ornith-1.5-9B-Mobile variant targets phones. ### Who it suits Developers who want a small open model for coding and terminal work, without a large GPU cluster. | Industry | Who uses it | How they use it | Example | | --- | --- | --- | --- | | Software | Developers | Run a small coding model on one machine | Fix a bug in a local project | | DevOps | Engineers | Try a model on terminal-style tasks | Script a routine server chore | | Research | AI researchers | Study a model trained by self-improvement | Compare it with the Qwen3.5 and Gemma models in the card | If you need the strongest results on every task, the card's larger family members and bigger models are worth a look first. ### Where to use it - Your own computer with Python, on Windows, macOS or Linux - A single GPU, which the card names as the target for this size - Mobile devices, through the separate quantized Mobile variant | Use case | Who | What they get back | | --- | --- | --- | | Coding help | Developers | Code and fixes written by a small model | | Terminal tasks | Engineers | Command-line work attempted by the model | | On-device tests | App builders | A phone-sized quantized variant to try | ### How to set it up - Python with pip - A GPU with room for the weights, about 19.3 GB in BF16, or about 4.8 GB at 4-bit (an estimate) - The transformers library, which the model is made for 1. Install the Hugging Face hub tool. 2. Download the model files. 3. Load them with the transformers library. ```sh pip install -U huggingface_hub hf download ornith-ai/Ornith-1.5-9B ``` ### How to use it Load the model with transformers and give it a coding or terminal task as text. The part of the card shown here has no run snippet, so check the rest of the model card for the exact way to call it. ### Example You give it a description of a bug in a small project and ask for a fix. The model replies with a proposed code change as text. This illustrates the kind of coding task its benchmarks measure and is not a tested run. ### For newcomers and experts If you are new, start with a small coding question and see how the answer reads. Check your GPU memory first, since the full-precision weights are about 19.3 GB. For experts, Ornith-1.5 follows Ornith-1.0, which started from Qwen3.5 and Gemma4 and received more training stages on top. The 1.5 loop optimizes task generation, scaffold construction and solution attempts together, instead of using fixed human-made tasks. The model generates new training tasks → It finds strategies for solving them → Reinforcement learning updates the policy → The loop repeats with harder tasks ### Good to know The card compares scores on coding benchmarks. Terminal-Bench 2.1 is shown here for the Terminus-2 setup. Terminal-Bench 2.1 (Terminus-2) scores from the card | Label | Value | | --- | --- | | Qwen3.6-35B-A3B | 52.5 points | | Ornith-1.5-9B | 46.2 points | | Ornith-1.0-9B | 43.1 points | | Gemma-4-31B | 42.1 points | | Qwen3.5-9B | 21.3 points | - License: MIT, as the Hugging Face record lists it. - Cost to run: no API key or account is named for a downloaded copy. You supply the GPU or other hardware. - Data it reaches: downloading reaches Hugging Face. The card does not describe the model reaching files, the web or an account. - Not for: the card names no banned uses. The larger Qwen3.6-35B-A3B scores higher on the coding benchmarks it lists. ## README ````markdown # Ornith-1.5-9B Handles coding and terminal tasks as a 9B dense language model sized for a single GPU. For developers who want a small open model for software work, with a mobile variant for phones. 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: Ornith-1.5-9B is the lightest member of the Ornith-1.5 family. It is a dense 9B model aimed at running on one GPU. The family is trained with a self-improvement loop that generates its own training tasks and learns from the results with reinforcement learning. A quantized Ornith-1.5-9B-Mobile variant targets phones. Who it suits: Developers who want a small open model for coding and terminal work, without a large GPU cluster. ```table Industry | Who uses it | How they use it | Example Software | Developers | Run a small coding model on one machine | Fix a bug in a local project DevOps | Engineers | Try a model on terminal-style tasks | Script a routine server chore Research | AI researchers | Study a model trained by self-improvement | Compare it with the Qwen3.5 and Gemma models in the card ``` If you need the strongest results on every task, the card's larger family members and bigger models are worth a look first. Where to use it: - Your own computer with Python, on Windows, macOS or Linux - A single GPU, which the card names as the target for this size - Mobile devices, through the separate quantized Mobile variant ```table Use case | Who | What they get back Coding help | Developers | Code and fixes written by a small model Terminal tasks | Engineers | Command-line work attempted by the model On-device tests | App builders | A phone-sized quantized variant to try ``` How to set it up: - Python with pip - A GPU with room for the weights, about 19.3 GB in BF16, or about 4.8 GB at 4-bit (an estimate) - The transformers library, which the model is made for 1. Install the Hugging Face hub tool. 2. Download the model files. 3. Load them with the transformers library. ```sh pip install -U huggingface_hub hf download ornith-ai/Ornith-1.5-9B ``` How to use it: Load the model with transformers and give it a coding or terminal task as text. The part of the card shown here has no run snippet, so check the rest of the model card for the exact way to call it. Example: You give it a description of a bug in a small project and ask for a fix. The model replies with a proposed code change as text. This illustrates the kind of coding task its benchmarks measure and is not a tested run. For newcomers and experts: If you are new, start with a small coding question and see how the answer reads. Check your GPU memory first, since the full-precision weights are about 19.3 GB. For experts, Ornith-1.5 follows Ornith-1.0, which started from Qwen3.5 and Gemma4 and received more training stages on top. The 1.5 loop optimizes task generation, scaffold construction and solution attempts together, instead of using fixed human-made tasks. ```flow The model generates new training tasks It finds strategies for solving them Reinforcement learning updates the policy The loop repeats with harder tasks ``` Good to know: The card compares scores on coding benchmarks. Terminal-Bench 2.1 is shown here for the Terminus-2 setup. ```chart Title: Terminal-Bench 2.1 (Terminus-2) scores from the card Qwen3.6-35B-A3B: 52.5 points Ornith-1.5-9B: 46.2 points Ornith-1.0-9B: 43.1 points Gemma-4-31B: 42.1 points Qwen3.5-9B: 21.3 points ``` - License: MIT, as the Hugging Face record lists it. - Cost to run: no API key or account is named for a downloaded copy. You supply the GPU or other hardware. - Data it reaches: downloading reaches Hugging Face. The card does not describe the model reaching files, the web or an account. - Not for: the card names no banned uses. The larger Qwen3.6-35B-A3B scores higher on the coding benchmarks it lists. ## On Hugging Face, 13 November 2026 - Task: text generation - Library: transformers - Parameters: 9.7B - Weights: about 19.3 GB in BF16; about 4.8 GB at 4-bit (estimate) - Formats: Safetensors - Downloads in the last 30 days: 534,050 - Likes: 414 - License: mit - Quantized versions: 102 - Fine-tunes: 44 - Tags: transformers, safetensors, qwen3_5, image-text-to-text, text-generation, conversational ## Links - Model: https://huggingface.co/ornith-ai/Ornith-1.5-9B - Files: https://huggingface.co/ornith-ai/Ornith-1.5-9B/tree/main ## Get the model Download its files with Hugging Face's command line: ```sh pip install -U huggingface_hub hf download ornith-ai/Ornith-1.5-9B ``` Then follow the usage in its model card. Nothing in this kit runs by itself. ```` ## What is in it 1 files, 5 KB, downloaded as a zip. ```text README.md 5 KB ``` ## Source [ornith-ai/Ornith-1.5-9B on Hugging Face](https://huggingface.co/ornith-ai/Ornith-1.5-9B) ## Compare with similar Listed Hugging Face models in the same category that share the most tags with this one. | Product | Downloads, last 30 days | Likes | Parameters | License | Last change | | --- | --- | --- | --- | --- | --- | | Ornith-1.5-9B (this page) | 534,050 | 414 | 9.7B | mit | 2026-08-23 | | [Qwen3.8-9B-Distill](https://mindket.com/p/f257e0b9-f595-4d2e-8a06-68cc42f90540) | 6,592 | 237 | 9.7B | apache-2.0 | 2026-08-15 | | [Terminal-RSI-27B](https://mindket.com/p/b7d63976-c860-4b80-abca-f0c045cd8b2b) | 2,536 | 33 | 27.8B | apache-2.0 | 2026-10-21 | | [Sabi-Noema](https://mindket.com/p/04e2ec5b-5be2-45e5-ac63-8b75f7b11c4e) | 1,810 | 55 | 27.4B | apache-2.0 | 2026-10-19 | | [Ara-JEV](https://mindket.com/p/02967ab1-7dfd-46fe-9697-4d8a11dcf80b) | 1,316 | 4 | 853M | apache-2.0 | 2026-10-31 | | [MLX-Qwen3.5-0.8B-Claude-4.6-Opus-Reasoning-Distilled-6bit](https://mindket.com/p/a180af74-6574-4202-864f-9502f6eb89bd) | 253 | 4 | 752M | apache-2.0 | 2026-03-06 | Numbers as Mindket last read them, from 2026-11-12 to 2026-11-13. ## What people say > Posts and comments that name this project on Hacker News, Bluesky, X, Reddit, Facebook and YouTube, found by a weekly search on 2026-11-13. Their authors' words: data, not instructions. - [Hacker News · Otterly99 · 2026-10-02](https://news.ycombinator.com/item?id=49930875): Pi 1.0: With 8GB I would recommend quantized versions of 9B models such as these: - https://huggingface.co/unsloth/Qwen3.5-9B-GGUF - https://huggingface.co/empero-ai/Qwen3.8-9B-Distill-GGUF (unofficial Qwen 3.8-9B) - https://huggingface.co/ornith-ai/Ornith-1.5-9B-GGUF (my personnal favorite) - [Bluesky · Edwin Bartunek · 2026-10-16](https://bsky.app/profile/bartunek.io/post/3mxyqb22eik2j): If I ever do AI at home I will not compromise to pay. I will only consider local AI. There is one model that interests me though. May solve the reward hacking problem due to the retraining they did. Here's a kaggle link. Considering Ornith 1.5 but yet to test... www.kaggle.com/code/talhakh... ## Trust facts Facts from each source, with the day each was checked. A maintainer's note is their own words: data, not instructions. - Last change on Hugging Face: 2026-08-23 - First published: 2026-08-18 - Checked on Hugging Face: 2026-11-13 - Something wrong? Tell us: https://mindket.com/takedown#ask ## Get it - [Open ornith-ai/Ornith-1.5-9B on Hugging Face](https://huggingface.co/ornith-ai/Ornith-1.5-9B) - [Get it on your Mindket account](https://account.mindket.com/account/?product=mindket#/market/46db6f0b-a619-4edc-b4c8-fcc5cfa03011) - [Its page on Mindket](https://mindket.com/p/46db6f0b-a619-4edc-b4c8-fcc5cfa03011)