What's the best model you have running on strix halo 128GB?
retained text
I have a framework desktop with 128gb of unified memory and have been having a blast learning local models with it. Ive noticed something, at least in my cases I always end up using smaller or moderately sized models like qwen3.6 35b or qwen 3.8 27b. Usually these models give me the best results for coding and im not complaining. But I feel like that extra 90GB is just sitting around being wasted, and the CPU isn't beefy enough to really run multiple agents at a decent speed. What have you all been running on your strix halo machines, and am I missing out on something?
document fetched 2026-09-17 00:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1wi5e8v/whats_the_best_model_you_have_running_on_strix