What's the alternative to frontier models? Disk-streamed GLM 5.1? By the time you get a single response back, the API will be back up.

35t/s on Qwen3.5 122b on a Strix Halo. The local stuff works great now. Stop giving the corpo monopolists money.