Flow - 4x faster than typing
Mini toolsBonsai 27B
By PrismML
Model family: Qwen
Bonsai 27B is PrismML's largest Bonsai release, a 27.8B-parameter multimodal model based on Qwen3.6 27B that accepts both images and text. The language model is quantized end-to-end to 1-bit binary or 1.58-bit ternary weights across embeddings, attention, MLPs, and the LM head, while its vision tower runs at 4-bit. It supports a 262,144-token context window and ships with thinking mode enabled by default, toggleable per request. It handles function calling and tool use through a standard OpenAI-compatible API, making it suitable for agentic workflows, coding, debugging, and multi-step planning. The 1-bit variant compresses to about 3.9GB, achieving over 90% of full-precision benchmark performance and running at 11 tokens/s on an iPhone 17 Pro Max. Released under the Apache 2.0 license.
New
Multimodal
Overview
Bonsai 27B is a 27.8B-parameter multimodal language model built on Qwen3.6 27B, quantized end-to-end to 1-bit or 1.58-bit ternary weights across embeddings, attention, MLPs, and the LM head. It accepts text and image input with a 262,144-token context window, and supports reasoning, tool use, and coding, while running locally on phones, laptops, and edge GPUs.
About PrismML
Industry:
Artificial Intelligence
Company Size:
6
Location:
Pasadena, California, US
Website:
prismml.com
Last updated: July 15, 2026
