Edge AI will eventually gain frontier AI capabilities. We have evidence. Manufacturing will get superpowers that only gigantic datacenters would possess. Not in “distant future”. A couple of years will be enough.
Yesterday we got shocked with the news of a new small model release from Alibaba. Fully open. Free. Local. Finetune it, change it, anything you wish.
Qwen3.8-27B is a good example of how fast local models are becoming actually useful.
It is a 27B dense model, yet in some coding and agentic benchmarks it gets close enough to much larger frontier models that comparisons with models like Opus 5 are no longer completely absurd.
The more interesting part is hardware.
Quantized, it can run fully offline on a high-end consumer laptop. A Ryzen AI Max+ system with 64–128 GB unified memory is already enough for this class of model.
That means no API dependency, no token pricing, and no requirement to send your code or data to somebody else’s servers.
It is obviously not “GPT 5.6 Sol but local.” That would be a stupid conclusion but how close they get in some tasks is just… nuts!
27B models entering this performance range at all is a pretty significant shift.
This is NOT the end. Many more models will come soon. Even more optimized.
At some point, we will finetune these and merge capabilities and that’s when big factories can run thousands of instances of frontier intelligence completely offline, hidden from the outside world.