unsloth/gpt-oss-120b-GGUF

Text Generation • 117B • Updated Aug 25 • 101k • 188

Note Runs very fast on my 128G MacBook, full context, unquantized (beyond the release), with plenty of room to spare. The smartest model I can run locally, and on par with leading closed-weight models. My current favorite model. Caveat: Highly censored.

unsloth/Qwen3-235B-A22B-GGUF

Text Generation • 235B • Updated Jun 8 • 13.3k • 70

Note Just barely fits on a 128G MacBook (IQ4_XS). On par with some leading closed-weight models. The second smartest model I can run locally, and yet still fast.

unsloth/gpt-oss-20b-GGUF

Text Generation • 21B • Updated 9 days ago • 148k • 519

Note The best local model for modest laptops and desktops (16G to 32G). Use llama.cpp's --cpu-moe to leverage the system's modest GPU to speed up inference. Caveat: Highly censored.

bartowski/Qwen_Qwen3-32B-GGUF

Text Generation • 33B • Updated Apr 28 • 1.92k • 38

Note Decent mid-sized local model for debugging and solving problems. Smarter than some ~70-120B models.

bartowski/Qwen_Qwen3-30B-A3B-GGUF

Text Generation • 31B • Updated Apr 29 • 7.19k • 57

Note A GPU-poor workstation (32G RAM) model. Load the active parameters on your GPU, run experts on the CPU (-ot exps=CPU), making the most of all your hardware. Weak at programming, but otherwise decent and very fast. (If you can run this, better to run gpt-oss-20b unless the censorship is an issue.)

HuggingFaceTB/SmolLM2-360M-Instruct

Text Generation • 0.4B • Updated Sep 22 • 108k • 169

Note My favorite model for cramped spaces. I can even run this on my old Windows XP laptop with 1G RAM. A fun toy, but not good for much.