My new favorite local neuronal pet net is qwen3-14b. I tested it for a while in different tasks on a 16gb M1.
It does perform somewhat ok as 4b version with longer context in openhands but it is more for playing around and being curious and stuff.
Best i could get out of my noobcoder without feeding small prompts was a flask hello world page and a small company page with some css styling.
It is much worse in this executive agent role than just outputting text directly. The openhands documentation mentions that local models are too small for this task.
Still fun.
#openhands #llm