Bonsai 27B (1-bit LLM): The First 27B-Class Model to Run on a Phone

PrismML has announced Bonsai 27B, the first 27-billion-parameter language model capable of running on a smartphone, achieving this through 1-bit quantization that reduces the model to just 3.9GB while retaining 90% of the full-precision baseline's performance. The model supports advanced capabilities including multi-step reasoning, tool calling, vision tasks, and agentic loops, with a ternary variant (5.9GB) offering 95% performance retention for laptops and a fully multimodal architecture with a 262K-token context window. Both variants are available under the Apache 2.0 License.