qalarc.com / projects / local-model-atlas

Local Model Atlas

A project by Qalarc — AI systems studio, Sydney, Australia. Built, shipped and maintained by Qalarc.

Measured catalogue of every local AI model the studio runs — LLMs, image/video and audio — with real sizes, speeds and use-notes from daily production work.

OllamaComfyUIONNXAMD ROCmUnified memory

Which local model should you actually use? The internet answers with launch-day listicles; this atlas answers with a working studio's installed set: every model on our machines, its real disk footprint, its measured speed on unified-memory hardware, and what we use it for.

Covers the Ollama language fleet (GLM 4.7 Flash at 38 tok/s, Qwen3.8, Ornith 1.5, Qwen2.5-Coder), the ComfyUI image and video stack (z-Image Turbo, FLUX.1, the Wan 2.2 family, LTX 2.3), and the audio studio (ACE-Step music, F5-TTS voice cloning, MelBand Roformer stems). Continuously re-verified against what is actually installed.

Local Model Atlas

Tags