I am impressed and I owe you one, Qwen 3.8 flash next (vision)!

I have enabled the vision for the CIRU Strix UL4 quant of Qwen 3.8 flash next (others quants likely perform very similar) and tried it on a few things, then wanted to show my partner how great it works and she asked it it could identify plants. So I took a photo from a plant that we recently got as a gift from family and Qwen not only accurately identified the plant as oleander (Nerium oleander) but also warned that it's poisonous and (among other warnings) that you should keep pets/children away. We have a kid and both of us didn't know! I verified the Qwen identification and the poisonous claim and both checked out as accurate. The plant will have to go, thank you Qwen!!!

Stoked by the precision of combining a decent vision model with the domain knowledge of a ~180B params model (including ngrams) to actually identify and reason about what it sees, I took a photo of a pre-diagnosed skin condition of myself and the Qwen diagnosis was highly accurate again! This model may be really useful if you want to check something on your private parts real quick without visiting a dermatologist, e.g., or sending pictures of yourself to a cloud service (EDIT: of course it's only a first step before you visit a professional if it isn't obviously harmless/treatable by yourself! Qwen Flash will suggest to visit a doctor anyways along its assessment).

PS.: Hardware Strix Halo Box, CIRU Strix UL4 llama-server fork and quants, Chatbox on iPhone as Chat with support to add photos to conversations.