Ternary Bonsai 27B Local Test | 27B Model Running on a Phone? | Reasoning, Coding, Game Dev

Name: Ternary Bonsai 27B Local Test | 27B Model Running on a Phone? | Reasoning, Coding, Game Dev
Uploaded: 2026-07-19T14:33:38.439Z
Duration: 11 min 40 s
Channel: Venelin Valkov
Description: Test of the Ternary Bonsai 27B model, a highly compressed multimodal AI model that can run on smartphones. The video compares the performance of the ternary variant with the Q4 quantization of Qwen 3.6 27B.

Venelin Valkov19 July 2026Watch on YouTube

Part of series

Ep. 8 · Coding Cpp Llama

View the series

Description

Bonsai 27B is a multimodal model by PrismML, based on Qwen 3.6 27B, extremely quantized in ternary or binary format. The authors claim that the ternary variant preserves ~95% of the intelligence of the full-precision Qwen model. We'll take the Ternary Bonsai 27B for a local test with llama.cpp and compare it to the Q4 version of Qwen 3.6 27B and verify the claims. Weights: https://huggingface.co/prism-ml/Ternary-Bonsai-27B-gguf Announcement: https://prismml.com/news/bonsai-27b AI Academy: https://mlexpert.io/ Work with me: https://mlexpert.io/consulting LinkedIn: https://www.linkedin.com/in/venelin-valkov/ Follow me on X: https://twitter.com/venelin_valkov Discord: https://discord.gg/UaNPxVD6tv Subscribe: http://bit.ly/venelin-subscribe GitHub repository: https://github.com/curiousily/AI-Bootcamp 👍 Don't Forget to Like, Comment, and Subscribe for More Tutorials! Join this channel to get access to the perks and support my work: https://www.youtube.com/channel/UCoW_WzQNJVAjxo4osNAxd_g/join

What you'll learn

Ternary Bonsai 27B is a heavily compressed multimodal AI model that can run on smartphones and local devices through extreme quantization
The model reportedly preserves approximately 95% of the intelligence of the full Qwen 3.6 27B model despite significant compression
The ternary variant can be tested and compared with other quantized versions like Q4 of Qwen 3.6 27B using llama.cpp

Frequently asked questions

What is Ternary Bonsai 27B and how does it differ from standard AI models?

Ternary Bonsai 27B is a multimodal model by PrismML based on Qwen 3.6 27B that is extremely compressed in ternary or binary format. This enables running the model on smartphones and local devices, which is not feasible with standard models.

How much of the original performance is retained with ternary compression?

According to the developers, the ternary variant preserves approximately 95% of the intelligence of the full Qwen 3.6 27B model. This is tested and compared in the video with the Q4 quantization of the same model.

What are the practical benefits of running AI models locally on a phone?

Running AI models locally on a phone offers benefits such as privacy, no internet connection required, and lower latency. This enables advanced AI applications for reasoning, coding and game development without dependency on cloud services.

Which tools and frameworks are used for testing Ternary Bonsai 27B?

The video uses llama.cpp for testing the Ternary Bonsai 27B model and compares it with the Q4 version of Qwen 3.6 27B. The model weights are available on Hugging Face.

Topics

model quantization edge ai local inference open-source models