---

license: apache-2.0

library_name: gguf

pipeline_tag: image-text-to-text

language:

• en
• zh

tags:

• zen5
• zen5-nano
• zen5-nano-9B
• zenlm
• hanzo
• multimodal
• vision-language
• edge
• mobile
• zen-5

---

Zen5 Nano 9B

Edge / mobile / on-device tier of the Zen5 family. Multimodal (image + text in → text out) dense model at the 10B-parameter scale, tuned for fast on-device inference on phones, laptops, and Raspberry Pi class hardware.

Part of the Zen5 nano sub-family — pick by hardware budget:

| SKU | Params | Fit |

|---|---|---|

| zen5-nano-0.8B | 0.9B | Raspberry Pi 5 / phone CPU / browser WASM |

| zen5-nano-2B | 2B | low-end iGPU / 8 GB RAM laptop |

| zen5-nano-4B | 5B | 16 GB RAM laptop / mobile NPU |

| zen5-nano-9B | 10B | 24 GB+ unified RAM / consumer GPU |

The larger tiers (zen5-flash, zen5, zen5-pro, zen5-max, zen5-coder) cover everything from agentic-default to frontier-quality multi-GPU.

Weights

GGUF mirror is staged. Until it lands here, the recommended path is the hosted zen5-nano-9B endpoint below, or any community 10B-class multimodal GGUF placed in a local model/ directory.

Run

Hosted via the Hanzo gateway (api.hanzo.ai) as zen5-nano-9B.

Local with llama.cpp or any GGUF runtime:

llama-cli -m model.gguf --mmproj mmproj.gguf -p "What is shown in this image?"