Bonsai 27B: First 27B-Class AI Model to Run on a Smartphone
Original: Bonsai 27B: A 27B-Class model that runs on a phone
Why This Matters
On-device deployment of 27B-class models could enable powerful agentic AI workloads without cloud dependency.
PrismML announced Bonsai 27B on July 14, 2026 — the first 27B-class multimodal model to run on a smartphone. Based on Qwen3.6 27B, it comes in two variants: a 5.9 GB ternary build and a 3.9 GB 1-bit build compatible with the iPhone 17 Pro. Both are released under Apache 2.0.
PrismML launched Bonsai 27B, a multimodal AI model based on Qwen3.6 27B, claiming it is the first model of its capability class to run on a phone. Standard 27B models require approximately 54 GB in 16-bit precision; even aggressive 4-bit builds weigh around 18 GB — too large for mobile devices. Bonsai 27B addresses this with extreme weight quantization across two variants. The Ternary Bonsai 27B uses {−1, 0, +1} weights at 1.71 effective bits per weight, resulting in a 5.9 GB file suitable for laptops. The 1-bit Bonsai 27B uses binary {−1, +1} weights at 1.125 effective bits per weight, compressing the model to 3.9 GB — fitting within the memory budget of an iPhone 17 Pro. Both variants are fully multimodal, support a 262K-token context, and include speculative decoding. The vision tower is delivered in 4-bit precision. On a 15-benchmark suite covering math, coding, tool-calling, and vision, the ternary variant retains 95% of full-precision performance and the 1-bit variant retains 90%. PrismML states the 1-bit build delivers an intelligence density of 0.53 per GB — more than 10x the full-precision baseline. All weights are available today under the Apache 2.0 License.