Alpie Seed 9B

Alpie Seed 9B

Reads text, images and video and handles reasoning and coding with a 9B open-weight model built on Qwen3.5-9B, with GGUF builds for light hardware. For developers who want an offline-capable model.

Made by 169Pi 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: Alpie Seed 9B is 169Pi AI's multimodal model, based on Qwen3.5-9B. Post-training with supervised fine-tuning and GRPO strengthened its reasoning, instruction following, coding and visual understanding. It takes text, image and video input, has a 262K native context, and ships in several quantized formats.

Who it suits: It suits developers who want a 9B open-weight model for reasoning and coding that can also run on laptops or phones through the smaller GGUF builds.

Industry | Who uses it | How they use it | Example
Software | Developers | Run a 9B reasoning and coding model locally | Ask for help with a coding task
Mobile | App builders | Run a quantized build offline on a phone | Answer questions with no cloud connection
Research | Teams working on Indic languages | Test a model with Indian-language support | Check answers in Hindi or Bengali

People who need image input through Ollama can skip the latest tag, since the card lists it as text input.

Where to use it:

• Hugging Face Transformers, for the main weights
• Ollama, through the alpie-seed library entry
• GGUF builds on Hugging Face and Kaggle, in 2-bit, 3-bit and 4-bit sizes
• Windows, macOS or Linux, where you can download the files with Python

Use case | Who | What they get back
Local reasoning and coding | Developers | Answers with thinking and non-thinking modes
Offline mobile use | App builders | A quantized model that runs without a cloud connection
Long documents | Analysts | A 262K native context window

How to set it up: You need these first:

• Python and the huggingface_hub package
• Room for the files, about 18.8 GB in BF16 or about 4.7 GB at 4-bit as an estimate
• Or Ollama, if you prefer the packaged build

1. Install the Hugging Face hub package and download the files.
2. Or install Ollama and run the model from its library.
3. Pick a quantized build if your hardware is small.

pip install -U huggingface_hub
hf download 169Pi/Alpie_seed

ollama run 169pi/alpie-seed

How to use it: Run the Ollama command to chat with the model, or load the weights in Transformers. The card text gives no Transformers code, so check its model card for that route.

Tool | What it does
ollama run | Starts the model from the Ollama library
alpie-seed:latest | The default Ollama tag, 5.6 GB with 256K context and text input
GGUF builds | 2-bit, 3-bit and 4-bit files for lighter hardware

Example: You run the Ollama command and ask a coding question. The card says the model supports thinking and non-thinking modes, so it can reason through the problem before it answers.

For newcomers and experts: A 9B model is small enough for a good laptop or a phone when it is quantized. A first small step is the Ollama command above, then a short question.

The model follows the Qwen3.5-9B architecture and goes through SFT and GRPO post-training. The card says its technical report covers reward design, quantization, offline results on real devices and safety evaluation. Besides FP16 weights, the repository offers GGUF files at three bit depths: 2-bit, 3-bit and 4-bit.

Start from the Qwen3.5-9B architecture
Post-train with SFT and GRPO
Publish FP16 weights and GGUF builds
Run in Transformers, Ollama or offline

Good to know:

• Size: 9.4B parameters per Hugging Face, about 18.8 GB in BF16 and about 4.7 GB at 4-bit, as an estimate
• Languages: English, Hindi, Bengali, Marathi, Gujarati, Telugu, Malayalam and Punjabi are listed
• Versions: Hugging Face lists 5 quantized versions made by other people

Title: Size in GB
BF16: 18.8 GB
Ollama latest tag: 5.6 GB
4-bit estimate: 4.7 GB

• License: apache-2.0
• Cost to run: none named, no API key or account
• Data it reaches: it downloads model files from Hugging Face or Ollama, and the card shows it running offline on a phone
• Not for: image input through the latest Ollama tag, which the card lists as text input

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 9.4B
• Weights: about 18.8 GB in BF16; about 4.7 GB at 4-bit (estimate)
• Formats: Safetensors, GGUF
• Downloads in the last 30 days: 741
• Likes: 26
• License: apache-2.0
• Languages: en, hi, bn, mr, gu, te, ml, pa
• Based on: Qwen/Qwen3.5-9B
• Quantized versions: 5
• Tags: transformers, safetensors, qwen3_5, image-text-to-text, alpie, 169pi, qwen3.5, vision-language, multimodal, reasoning, coding, video-understanding

Links

• Model: https://huggingface.co/169Pi/Alpie_seed
• Files: https://huggingface.co/169Pi/Alpie_seed/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download 169Pi/Alpie_seed

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