PrismML Releases Bonsai 27B, First 27B AI Model Compressed to 3.9 GB for iPhone

According to PrismML, the company released Bonsai 27B, a 27-billion-parameter AI model compressed to 3.9 GB, earlier this week—small enough to run on an iPhone 17 Pro Max at 11 tokens per second. The ternary variant retains 94.6% of full-precision benchmark performance, outperforming conventional low-bit Qwen builds that are nearly twice as large and collapse on math and coding tasks. The binary variant reaches 1.125 bits per weight by quantizing model parameters to single binary signs with shared scaling factors.

Apple is in early talks with PrismML about the compression technology, according to CNBC. PrismML CEO Babak Hassibi confirmed the company is evaluating the technology for potential on-device use. A compressed Gemma model is next in the pipeline, followed by larger frontier models; 1-bit Bonsai 27B is available for free download under Apache 2.0.