I’ve been saying this for over a year now: Frontier models are fantastic, but the real future is frontier-level models (in every way) runnin…

I’ve been saying this for over a year now: Frontier models are fantastic, but the real future is frontier-level models (in every way) running locally on your own hardware. I think this is 18-24 months away at most. There's already been significant progress made with Gemma 4, Qwen 3.6 27B etc. but these models will still hit bottlenecks, make serious mistakes & can freeze up even on elite hardware. (I put Gemma 4 through it's paces, no where near Opus 4.7 for example). Current-gen local models are great, but not truly frontier level. Not yet. Vector quantization, architecture advances & leaps in personal computer power (I'm looking at you @Apple Silicon) are happening fast. Soon we will have local parity, lower long term cost, near-zero latency, with full privacy. This opens up the door for the distribution & decentralization of compute I've been discussing as well.

Source: Clem Delangue (X) | 2026-04-28