600tok/s single request on qwen3.6 35ba3b with Ninfer on an RTX Pro 6000. Anybody remember that Comcast ad "stupid fast"?

It's not the brightest bulb but it's my new drudgework model for read+find or code tasks I'm willing to let it brute force. Even if it takes 20x more tokens, that's still faster than many local models. Not quite Cerebras but still pretty fun to drive.