Qwen3.5-9B Brainwaves DeepThought STEM OneJev1.6

Qwen3.5-9B Brainwaves DeepThought STEM OneJev1.6

An experimental 9.4B image-and-text model made by nightmedia, a merge of three 9B models aimed at reasoning and science questions. It suits hobbyists and researchers who like to try small open merges.

Made by nightmedia 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 is an experimental merge made by nightmedia, an independent AI lab that says it works from a single laptop. It joins three models: its own Qwen3.5-9B-Brainwaves, xtools-at's Qwen3.5-9B-DeepThought-STEM and OmniJev's OneJev1.6-9B. The result has about 9.4 billion parameters, takes text and images, and lists English, Chinese, Japanese and Spanish. The card is more of a lab notebook than a spec sheet, so treat it as an experiment.

Who it suits: It suits hobbyists and researchers who enjoy testing small open merges, and people who want to compare how a merge scores against its parts. The model is little used so far, with 30 downloads in the last 30 days.

Title: Who might use it
Industry | Who uses it | How they use it | Example
Research | Model tinkerers | Compare a merge with the models it came from | Check its ARC score against the Brainwaves model
Local AI | Hobbyists | Try a small reasoning-minded model at home | Ask it a science question in English or Chinese
Education | Students of model merging | Study how combining models changes scores | Read the card's score tables side by side

If you need a tested, supported model for production, you can skip it. The card calls it experimental and gives no safety or limits section.

Where to use it: The record lists the transformers library and an image-text-to-text task. The card shows a test chat with the model, and gives no loading code in the visible part.

• Python with the transformers library
• Local experiments on a machine with enough memory for 18.9 GB of BF16 weights
• Windows, macOS or Linux with Python, to download the files

Title: Jobs it does
Use case | Who | What they get back
Reasoning and science questions | Hobbyists | Long written answers on technical topics
Image and text chat | Experimenters | Replies that take a picture and a prompt
Merge comparison | Researchers | Scores to set against its parent models

How to set it up: You need Python with pip and memory for about 18.9 GB of BF16 weights, or about 4.7 GB at 4-bit by estimate. The visible part of the card gives the download step and no run commands, so look at the transformers guide for an image-text model.

• Python with pip
• The transformers library
• Memory for about 18.9 GB of BF16 weights, or about 4.7 GB at 4-bit

1. Install the Hugging Face download tool.
2. Download the model files to your computer.
3. Load them with transformers, as you would another image-text model.

pip install -U huggingface_hub
hf download nightmedia/Qwen3.5-9B-Brainwaves-DeepThought-STEM-OneJev1.6

How to use it: Prompt it in chat form. The card's own test asks for a long analysis and a self-reflection, and the model answers with a long, structured report. Expect to experiment with prompts and check its answers, since the card gives no recommended settings.

Example: You send it a question about how attention in a transformer compares with ideas from physics. The card shows the model replying with a long report with headings and equations. Check the claims in a reply like that before you use them.

For newcomers and experts: If you are new, a merge blends several models into one without training from scratch. Treat this one as a curiosity and compare it with its parents on your own questions.

If you know the field, the card reports its own benchmark scores in a compact table. The merge scores 0.701 on ARC at q8-hi, above the 0.686 of the Brainwaves component in BF16 and the 0.561 of the plain Qwen3.5-9B instruct model. These are the author's numbers, run on quantized files, so rerun them on your own tasks.

Text or an image goes in
The merged 9.4B model reads it
It writes a long reply
You check the reply yourself

Good to know: The card's ARC scores show the merge ahead of its parts. Higher is better.

Title: ARC score, q8-hi unless noted
This merge: 0.701
Brainwaves, BF16: 0.686
DeepThought-STEM: 0.570
Qwen3.5-9B instruct: 0.561

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the model files from Hugging Face, and the text and images you send it
• Not for: production use without your own tests, or tasks that need verified answers

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 9.4B
• Weights: about 18.9 GB in BF16; about 4.7 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 57
• Likes: 5
• License: apache-2.0
• Languages: en, zh, ja, es
• Based on: OmniJev/OneJev1.6-9B
• Tags: transformers, safetensors, qwen3_5, image-text-to-text, qwen3.5, bf16, distillation, coder-flash, claude, polaris, polaris-alpha, reasoning

Links

• Model: https://huggingface.co/nightmedia/Qwen3.5-9B-Brainwaves-DeepThought-STEM-OneJev1.6
• Files: https://huggingface.co/nightmedia/Qwen3.5-9B-Brainwaves-DeepThought-STEM-OneJev1.6/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download nightmedia/Qwen3.5-9B-Brainwaves-DeepThought-STEM-OneJev1.6

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