Frontier-agentic tier of the Zen5 family at the lowest cost in the lineup. Zen agentic MoE with ~10B active parameters per token; trained on large-scale real-world environment RL for tool use, code, and multi-step reasoning.

Repackaged from MiniMaxAI/MiniMax-M2.5 (modified-MIT, MiniMax) — quantized to GGUF from the abliterated variant by huihui-ai. Not trained from scratch — a permissively-licensed redistribution for the OSS-clean Zen model line.

Part of the canonical Zen5 ladder:

SKU	Hardware fit	This repo
`zen5-flash`	anything (4 GB VRAM)	zen-5-flash-gguf
`zen5-mini`	32 GB unified RAM (Q4_K_M)	← you are here
`zen5` (default)	24 GB+ VRAM	zen-5-gguf
`zen5-pro`	Mac M4 Max / DGX Spark / H100 80GB	zen-5-pro-gguf
`zen5-max`	Mac Studio M3 Ultra 512GB / 8x H100	zen-5-max-gguf

Files

File	Size	Quant
main GGUF (`*-Q4_K_M.gguf`)	~140 GB	Q4_K_M

Run

Hosted via the Hanzo gateway (api.hanzo.ai) as zen5-mini.

Local with llama.cpp or compatible:

hf download zenlm/zen-5-mini-gguf --local-dir gguf
MAIN=$(ls gguf/*-Q4_K_M.gguf | head -1)
llama-cli -m "$MAIN" -p "Reply in one sentence: what year is it?"

License

Upstream: MiniMaxAI/MiniMax-M2.5 by MiniMax, licensed modified-MIT (see the upstream LICENSE); abliterated variant by huihui-ai. This repository redistributes a quantized derivative; the upstream license terms are retained.

zenlm/zen5-mini-gguf

Zen5 Mini

Files

Run

License

zenlm/
zen5-mini-gguf