A 27 billion parameter AI model just got small enough to live on your phone. That sentence alone should stop you mid-scroll.
The team behind it is PrismML, and the model is called Bonsai 2.
They took Alibaba's Qwen3.8 27B, a model normally built for beefy cloud servers, and compressed it down to just 5.9 GB.
That's not a typo. A model that size usually needs serious compute to run, and now it fits on a smartphone.
Here's the part that actually matters: they kept 98.2% of the original benchmark performance. Barely any intelligence lost in the shrink.
Think of it like vacuum-sealing a winter coat into a tiny bag, except the coat still keeps you just as warm when you unpack it.
This unlocks real on-device AI. No internet dependency, no sending your data off to a server, no latency while you wait for a response.
Edge AI just got a serious upgrade, and the gap between 'cloud-only AI' and 'AI that lives in your pocket' is closing fast.
🔗 Original Source & Reference: https://www.techmeme.com/260917/p45#a260917p45
Published automatically via FeedMind AI Content Pipeline.
Top comments (0)