I strictly use only local models, so I agree, but the project was built with Fable so my argument hinges on the assumption that the maintainer is going to continue using Fable and needs to pay for it.
Which model works well?
Qwen 3.6 27b at q6 works decently well on 24GB of VRAM