GitHub radar
NeoHorse-1-9B: agent-ready 9B model for your laptop
TokenRhythm's NeoHorse-1-9B is a Qwen3.5-9B-based model trained for agent workflows, tool use, and coding — and it fits on a 16 GB laptop in quantized form.
NeoHorse-1-9B is TokenRhythm's model post-trained on top of Qwen3.5-9B, built for work in agent workflows. It can call tools and write code, and it follows complex instructions. In testing across ten benchmarks, it reaches a macro average of 69.04 compared to 65.60 for the base Qwen3.5-9B. The native context window is 262,144 tokens, which is enough to load a small project into one session. Quantized GGUF versions are available for use with Ollama or LM Studio. In 4-bit quantization, the model takes approximately 5.4 GB and runs on a laptop with 16 GB of RAM. For complex agent work, it does not replace a frontier model — but the quality is honest for a model of this size.
Why a vibe-coder should care
If you run local agents for simple coding tasks or automation, this fits on a standard laptop without extra hardware. The 262K context is enough to drop a small project into one session without splitting it. The quality is honest for a 9B model — not a replacement for Claude or GPT, but a working option for offline repetitive work.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install the NeoHorse-1-9B model from TokenRhythm locally: https://huggingface.co/TokenRhythm/NeoHorse-1-9B — find the right GGUF version for my machine, install via Ollama, and show me how to use it for coding tasks.
A regular laptop with 16 GB of RAM — the 4-bit version takes approximately 5.4 GB.
Open on Hugging Face▌ More finds