How much memory do the Ternary Bonsai 2 models use?

Ternary Bonsai 2 2B uses 0.43GB, Ternary Bonsai 2 4B uses 0.95GB, and Ternary Bonsai 2 9B uses 2GB. That is under one-ninth of their Qwen3.5 base models.

How can I try these models on a phone?

The free app Locally AI already supports running each model on iPhone and iPad. Update the app to try them.

Do the models lose accuracy compared with their base models?

PrismML says the drop from the base models is kept small, and on some tests the new models score higher than the base models.

Large Language ModelsOpen-Source AIGIGAZINE AI3h agoOct 16, 2026, 13:00 JST

PrismML's Ternary Bonsai 2 2B runs on 0.43GB as open models

Send on LINE

3 Key Points

What happened
PrismML released Ternary Bonsai 2 2B, 4B and 9B on October 15, 2026, remade from Qwen3.5 to process in three values, with memory of 0.43GB, 0.95GB and 2GB.
Why it matters
The models' memory use is kept to under one-ninth of their Qwen3.5 base models, while benchmark scores stay close to the base models or beat them on some tests.

WHO IT HITSApp developers and product teams building on-device AI assistants can now ship models that fit phones without a cloud connection, using the already-supported Locally AI app as a test path.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

PrismML built Ternary Bonsai 2 on Alibaba's Qwen3.5 family, then remade each model so it processes in three values, -1, 0 and 1, instead of the usual 16-bit or 4-bit formats. That change is what shrinks memory use to under one-ninth of the base models, with the 2B version landing at 0.43GB. The company designed the models around four behaviors for a local assistant: following instructions, avoiding processing loops, calling the right tools, and skipping unnecessary output.

PrismML also shared benchmark tables showing the new models stay close to their base models and beat them on some tests. "Locally AI," a free app, already runs all three, including a version of the 2B model shown on an iPhone 18 Pro by the app's developer, Adrien Grondin. It remains to be seen how these small three-value models hold up in wider daily use.

FAQ

How much memory do the Ternary Bonsai 2 models use?: Ternary Bonsai 2 2B uses 0.43GB, Ternary Bonsai 2 4B uses 0.95GB, and Ternary Bonsai 2 9B uses 2GB. That is under one-ninth of their Qwen3.5 base models.
How can I try these models on a phone?: The free app Locally AI already supports running each model on iPhone and iPad. Update the app to try them.
Do the models lose accuracy compared with their base models?: PrismML says the drop from the base models is kept small, and on some tests the new models score higher than the base models.

GIGAZINE AIRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.