I am currently using Llama-3.2-1B-Instruct-q4f16_1-MLC  via WebLLM v0.2.82. This is a completely local feature for making a personalised meal plan for the user as per their diet goal, even without the internet so they don't need to look at emails and other notifications first thing in the morning when they want a breakfast for say vegan meal for heart health. Llama works fine for this but anything a little deep in the conversation and its starts to become strange. I was thinking about qwen 3.5 0.8b, but would love to hear from you all, given you would have more experience.