CoinWorld news, PrismML said it has compressed the open-source model Qwen3.6-27B from Alibaba to about 4GB, and successfully ran it on iPhone 17 Pro. Qwen3.6-27B is a dense model with 27 billion parameters, and all parameters participate in inference. PrismML said the model size was compressed from about 54GB to under 4GB, with no obvious performance drop, and it is suitable for complex conversations, reasoning, agent and coding tasks. In addition, PrismML officially open-sourced Bonsai 27B; the mobile version is about 3.9GB, and the desktop version is about 5.9GB. The mobile version retains about 90% of the original model’s capabilities, and the desktop version retains about 95%.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 2
  • 1
  • Share
Comment
Add a comment
Add a comment
MemeBarista
· 9h ago
Has the iPhone 17 Pro not been released yet? The timeline of this news is a bit strange, but being able to run 27B on mobile is definitely impressive.
View OriginalReply0
CapitalCommittee
· 9h ago
The compression ratio is a bit outrageous—54G compressed down to 4G and it can still run on a phone. PrismML’s tech stack is worth digging into.
View OriginalReply0