By "open models", do you perhaps mean the approximately trillion-parameter open models that have been getting released recently? Because basically nobody can actually self-host those. And even at that size it's debatable that they're closing the gap.

Talk to me when you can do it with a 24B model quantized to 8 bits.