PrismML releases Bonsai 27B, a 1-bit model for laptops and phones

PrismML has released Bonsai 27B, a highly compressed version of Qwen3.6-27B designed for efficient operation on consumer hardware. The model features ternary and 1-bit weight representations, reducing its size to 5.9GB and 3.9GB respectively, while retaining a significant portion of the original model's performance. These low-bit builds are multimodal and support a large context window, making them suitable for devices with limited memory such as laptops and phones. AI

IMPACT Enables running large language models on consumer devices with limited memory, potentially broadening access and use cases.

RANK_REASON Release of a novel, highly compressed model variant with performance metrics. [lever_c_demoted from research: ic=1 ai=1.0]

Read on MarkTechPost →

• AIME26
• Apache 2.0
• Apple Silicon
• BitNet
• Bonsai 27B
• DSpark
• EvalScope
• Gemma-4-31B
• FP16
• HQQ
• iOS
• LiveCodeBench
• MATH-500
• MMLU-Redux
• OpenAI
• PrismML
• Qwen3.6-27B
• vLLM

• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →