Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint

PrismML released Ternary Bonsai 2 27B, a compressed 27B-parameter multimodal model that achieves a 9x smaller footprint (5.9GB) while retaining 98.2% of its full-precision counterpart's performance through ternary weight quantization. The model, based on Qwen3.8 27B, supports a 262K-token context window and multimodal inputs, enabling efficient local deployment with improved reasoning, coding, vision, and agentic capabilities compared to the previous Bonsai generation.