Biggest, baddest model to fill 144GB VRAM + 120GB RAM to the brim, regardless of speed

I'm trying to round out my quiver of daily driver models for my personal harness. Right now I drive qwen3.6 27b for balanced code and gemma4 31b for human interaction with lots of context and a few parallel sessions. Minimax M2.7 at Q6 clocks in at 207gb base and just barely fits once I get KV cache and context down for when I have a "take all day to answer; just be right" problem. I'm debating on moving to M3 at Q3, but I'm wondering if there are any other chonky models that will fill my 264GB with base + KV + context -- qwen3.6 is pretty special in terms of punching above its weight but I really want the most intelligent model possible for more complex reasoning, coding, and tool calling. Any favorites? Anyone compared M3@Q3 vs M2.7@Q6? They seem fairly equivalent to me but I love me some anecdata :)