← Back to Guides

Bonsai 27B: Complete Guide to Running a 27B-Class AI Model Free on Your Phone - Based on Qwen3.6, Apache 2.0 Licensed

PrismML introduces revolutionary Bonsai 27B, compressing 27B-parameter AI to 3.9GB for local iPhone running. Based on Qwen3.6, Apache 2.0 licensed, completely free.

Bonsai 27B: Complete Guide to Running a 27B-Class AI Model Free on Your Phone - Based on Qwen3.6, Apache 2.0 Licensed

Bonsai 27B is a low-bit quantized version of Qwen3.6-27B by PrismML, runnable on phones and laptops, Apache 2.0 licensed free.


What is Bonsai 27B

Bonsai 27B by PrismML is a low-bit quantized version of Qwen3.6-27B, not a newly trained model.

Key features:

  • Based on Qwen3.6 β€” Same architecture, just quantized
  • 5.9GB β€” Extremely small model file
  • Apache 2.0 β€” Completely free for commercial use
  • Runs on phones β€” 27B model on smartphones

How to Run

Option 1: On phone

  • Download PrismML's mobile inference engine
  • Install Bonsai 27B model
  • Run directly

Option 2: On laptop

  • Needs 8GB+ RAM
  • Use llama.cpp or similar engine
  • Model file only 5.9GB

Option 3: On server

  • Needs 14GB+ VRAM GPU
  • Better performance

Comparison with Original Qwen3.6

ModelSizePrecisionDeviceFree
Qwen3.6-27B~54GBFP16Serverβœ…
Bonsai 27B (1-bit)~5.9GB1-bitPhone/Laptopβœ…
Bonsai 27B (ternary)~8GBTernaryLaptopβœ…

Bonsai 27B shrinks Qwen3.6's model size by 9x while retaining most capabilities.


Who Should Use This

  • Phone AI enthusiasts β€” Run 27B model on smartphone
  • Laptop developers β€” No server needed
  • Privacy-conscious β€” Fully offline
  • Budget-conscious β€” Apache 2.0 free

Summary

Bonsai 27B is a quantized version of Qwen3.6, only 5.9GB, runs on phones and laptops, Apache 2.0 open-source free.

Project: prismml.com


Data current as of July 14, 2026. Bonsai is an open-source project, actively maintained.