[Etheryte]

You can already do that, just how slow or fast you go depends on how much you're ready to pay for memory. It's a $1200 premium to go from 36GB to 128GB of unified memory, that cost is hard to justify unless you really need it, or if someone else is paying.

[willahmad]

None is comparable to GPT-5 or Sonnet 4.5 experience

[elzbardico]

Frankly, right now I am way more satisfied with qwen-3-coder-420 using Cerebras inference than with those more powerful models.

Inference speed and fast feedback matter a lot more than perfect generation to me.

[mertbio]

Yet.