Back
RCreddit.com
14
·7 hr ago·RSS
Not on the current live radar

Models to download for M5 Ultra 512GB

View original
DeepSeekModel releasePlans & limits

Heat trend

Collecting trend data

The percentage is based on available heat signal, not comment count or independent people.

Why it matters

DeepSeek model activity is surfacing — worth tracking for capability changes, ecosystem impact, and availability.

AI summary

A user is preparing for an M5 Ultra 512GB setup with a TB5 enclosure and 4TB SSD, seeking advice on optimal model quants for the 512GB RAM constraint. They plan to download GLM-5.3-MLX-mixed-4_8bit (427.8GB), GLM-5.3-Flash-MLX-mixed-4_8bit (181.9GB), Qwen3.8-Flash-Next-MLX-mixed-4_8bit (106.2GB), DeepSeek-V4-Flash-MLX-mixed-4_8bit (165GB), and Kimi-K3-REAP73-MLX-mxfp4-q8 (451GB). They also inquire about context lengths for GLM-5.3 and Kimi-K3, and the viability of running 8-bit models.