# Tencent releases GGUF‑quantized Hy‑Memory‑30B‑A3B model on Hugging Face

> Three quantized versions (Q6_K, Q8_0, Q4_K_M) are available for local inference with llama.cpp, enabling efficient long‑term memory writing.

Oossa · 2026-10-19 · https://oossa.com/en/tencent-releases-gguf-quantized-hy-memory-30b-a3b-model-on-hugging-face

Tencent Hunyuan has added GGUF‑quantized files of its Hy‑Memory‑30B‑A3B model to Hugging Face. The repo offers three sizes—24.7 GB (Q6_K), 32 GB (Q8_0) and 18.2 GB (Q4_K_M)—all passing llama.cpp checks. The model acts as a memory writer inside frameworks like Mem0 and MemOS, deciding what information to keep and how to update it. In the vendor’s benchmark, the Q6_K version achieved the highest quality score, while the Q4_K_M version is the smallest download.

## The facts

- Three quantization levels: Q6_K (24.71 GB), Q8_0 (31.99 GB), Q4_K_M (18.24 GB).
- Published on Hugging Face on 2026‑10‑19.

## Why it matters

Developers can run a specialized long‑term memory model locally without needing cloud resources, saving bandwidth and hardware costs.

## Sources & references

1. [New model on Hugging Face: tencent/Hy-Memory-30B-A3B-GGUF (text-generation)](https://huggingface.co/tencent/Hy-Memory-30B-A3B-GGUF) – Tencent Hunyuan, 2026-10-19

Last updated: 2026-10-19