NeoHorse-1-9B targets agentic coding tasks

inclusionAI has released NeoHorse-1-9B, a dense 9-billion-parameter language model aimed at agentic tool use and coding. Published on Hugging Face under the TokenRhythm namespace, the model is described as being built on the Qwen3.5 foundation and tuned for the kinds of multi-step, tool-calling workflows that increasingly define practical developer tooling. You can find it at its Hugging Face page.

The 9B size puts NeoHorse-1 in a familiar sweet spot for teams that want a capable model they can run without the infrastructure demands of larger systems. Rather than chasing raw scale, this release leans into a specialization: agentic behavior and code, the two areas where smaller models have shown the most consistent gains over the past year.

Why it matters

Agentic coding assistants live or die by reliability, not just fluency. A model that reasons through a task and calls the right tools in the right order can outperform a much larger general-purpose model on real work. Positioning a 9B model specifically for this niche is a bet that focused tuning beats size for these jobs.

• Dense 9B model (not a mixture-of-experts design)
• Built on the Qwen3.5 base
• Focused on tool use, reasoning, and code
• Released under a custom ("other") license

A few practical details remain unspecified in the release record, including context length and formal license terms, so teams evaluating NeoHorse-1-9B for production should confirm those directly on the model card before committing.