llama3:8b, I know it's "far from ideal" but only really specific use cases require more advanced models to run locally, if you do software development, graphic design or video editing 8gb is enough
edit: just tried it after some time and it works better than I remembered showcase