NeoHorse-1-9B

NeoHorse-1-9B

A 9B text model post-trained from Qwen3.5-9B for agent frameworks, tool use, coding and instruction following, released as text-only weights. For developers who run agents on their own hardware.

Made by TokenRhythm 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: NeoHorse-1-9B is a text model with 9B parameters, released by TokenRhythm. They took Qwen3.5-9B and trained it further for agents that work in text, for calling tools, for coding and for following instructions. They describe it as a first prototype in work aimed at recursive self-improvement. The release holds language-model weights only.

Who it suits: Developers who build agents, coding helpers or tool-using systems and want a mid-sized open model they can run themselves.

Industry | Who uses it | How they use it | Example
Software | Developers of coding helpers | Generate and revise code | Ask for a function and a fix for a bug
Agent systems | Builders of text-based agents | Run the model inside an agent framework | Let it call tools and read the results
Research | People studying self-improvement | Test an early prototype for the idea | Compare it against its Qwen3.5-9B base

You can skip it if you need image input, since the vision weights are not included.

Where to use it:

• Python with the transformers library
• Text-based agent frameworks
• Windows, macOS or Linux, wherever you can run a 9B model

Use case | Who | What they get back
Tool use in agents | Agent builders | Text that drives tool calls
Coding help | Developers | Code and explanations as text
Instruction following | Anyone | Text replies that follow your instructions

How to set it up: You need Python and a GPU that can hold a 9B model. The weights are about 17.9 GB in BF16 and about 4.5 GB at 4-bit by estimate. The part of the card used here has no loading code, so check the model page.

• Python and pip
• A GPU with room for the weights
• The transformers library

1. Download the model files.
2. Open the model page for loading steps.
3. Load the model with transformers.

pip install -U huggingface_hub
hf download TokenRhythm/NeoHorse-1-9B

How to use it: Load the model as a text-only language model and use it inside an agent framework or a plain chat loop. The card says the weights were repackaged for text-only inference, with configuration and tensor key names changed but the fine-tuned values left as they were.

Example: You connect the model to a coding agent framework and ask it to fix a failing test. It returns text, such as a patch or a tool call. The card shows no sample output in the part used here.

For newcomers and experts: If you are new, start with a short instruction in a simple chat loop to see how it answers, before you wire it into an agent framework.

For experts, the model is derived from Qwen3.5-9B and fine-tuned for agentic post-training with a routing framework, as the card's title says. Repackaging drops the vision weights and changes tensor key names, so tools that expect the original Qwen3.5 layout may need adjusting.

Qwen3.5-9B base
Agentic post-training
Text-only repackaged weights
Agent framework or chat

Good to know:

• Size: 9B parameters, about 17.9 GB in BF16
• Text only: vision weights are not included
• Status: the authors call it an initial prototype
• License: Apache 2.0
• Cost to run: no API key or paid service named
• Data it reaches: downloads files from Hugging Face, nothing else is named
• Not for: image input, since the vision weights are left out

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 9B
• Weights: about 17.9 GB in BF16; about 4.5 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 1,548
• Likes: 1,051
• License: apache-2.0
• Based on: Qwen/Qwen3.5-9B
• Quantized versions: 15
• Fine-tunes: 5
• Tags: transformers, safetensors, qwen3_5_text, text-generation, agentic, tool-use, coding, reasoning, instruction-following, conversational

Links

• Model: https://huggingface.co/TokenRhythm/NeoHorse-1-9B
• Files: https://huggingface.co/TokenRhythm/NeoHorse-1-9B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download TokenRhythm/NeoHorse-1-9B

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