Current best model for narrative, chat, prompt creation (so basically everything except agentic coding)? - 5090

Im looking to set up a new local llm (probably on unsloth studio as that seemed to be doing pretty well last time I tested it). This one won't need to do agentic coding or app building or anything (not this time) but instead more 'text' based tasks such as - being given reference images and instructions, in order to then generate prompts for comfyui being sent long-form writing (paragraphs and chapters) and being able to sense-check and give feedback and rewrites. conversational help and problem solving (much like what I use chatgpt for atm). I would think qwen 3.8 might be the current best, but theres only a 27b model for it so its not really fast even on my hardware. The qwen3.6 35b would be a lot faster but I dont know how much response-quality im giving up for that. Figured id ask in case, for my purposes, theres actually a much better third option. Ive heard theres one called Ornith1.5 which seems to be a qwen3.6 fine tune, but i havent looked to see what its been tuned towards. If its more coding based then it won't help. Thanks!