Bonsai – A Competitive Ternary Weight LLM
Introducing Bonsai 0.5B, one of the first ternary-weight LLMs to rival full-precision models of similar size, such as Qwen 2.5 0.5B and MobileLLM 0.5B. Trained on just 3.8B tokens, using 1,000x less data than other models, Bonsai redefines what’s possible for ultra-efficient training in low-bit models. Next, we're building larger and more powerful ternary-weight models for the edge. Technical Report: https://github.com/deepgrove-ai/Bonsai/blob/main/paper/Bonsa... Model (Unpacked): https://huggingface.co/deepgrove/Bonsai Reach us:…
Bonsai – A Competitive Ternary Weight LLM launched on March 21, 2025 with 11 votes, #2,427 of 7,215 launches that month and more than 75% of that year's launches. The site was still online in September 2026, 1 year on.
In plain words
Bonsai is a 0.5-billion parameter language model using ternary weights, designed to run efficiently on edge devices. It matches the performance of similar full-precision models while trained on just 3.8 billion tokens, using significantly less training data than comparable systems. The model is built for developers and organizations needing lightweight language models that maintain competitive capabilities without requiring extensive computational resources.
written from the facts on this page · September 2026
From the sources
Contribute to deepgrove-ai/Bonsai development by creating an account on GitHub.
Bonsai is a small 500 million parameter ternary weight language model trained by deepgrove. Bonsai adopts the Llama architecture and Mistral tokenizer following Danube 3 , with modified linear layers to support ternary weights. The model has been trained primarily using DCLM-Pro and Fineweb-Edu. Bonsai marks a new paradigm of efficiency, being trained in less than 5 billion tokens. Bonsai achieves competitive performance among its peers, being one of the first ternary models to do so. Evalution results are below; for more detailed results and comparisons to other ternary models, please see the accompanying paper linked above. We use lm-eval for all benchmarks outside of MMLU and lighteval's…from github.com
In the maker’s words, at launch
Introducing Bonsai 0.5B, one of the first ternary-weight LLMs to rival full-precision models of similar size, such as Qwen 2.5 0.5B and MobileLLM 0.5B. Trained on just 3.8B tokens, using 1,000x less data than other models, Bonsai redefines what’s possible for ultra-efficient training in low-bit models. Next, we're…
Does a similar job
all alternatives →
1-Bit Bonsai, the First Commercially Viable 1-Bit LLMsMar 2026 · prismml.com · ▲430PrismML builds 1-bit and ternary Bonsai models that pack advanced reasoning, tool use, and multimodal AI into phones, laptops, and efficient data centers.
Bonsai 1.7B ternary model at 442T/s on M4 MaxMay 2026 · agents2agents.ai · ▲13We took a recently released Bonsai 1.7B ternary model from PrismML (https://github.com/PrismML-Eng/Bonsai-demo) and ran our agentic evolution search on it for 6 hours to optimize the Metal kernels. The search was fully autonomous. Measured against unmodified upstream llama.cpp at the same Bonsai/Q2_0 commit, same M4 Max: - tg128: 309.82 → 442.42 t/s (+42.0%) - pp512: 4250.32 → 4622.63 t/s (+8.8%)
- WBWeb browser to help programmers think clearly2021 · bonsaibrowser.com · ▲1,070
Data Bonsai: a Python package to clean your data with LLMs2024 · github.com · ▲47I've been doing some data cleaning for my fine tuning projects using LLMs, and decided to just build a package for it as a side project. Check it out here: https://github.com/databonsai/databonsai Some features: - categorization (labelling), transformation and decomposition (text into structured format) - validates llm outputs - batch mode batches up the inputs/outputs so you don't send the prompt (schema, fewshot examples) for every row of data, saving a significant amount of tokens There are some similarities to the Instructor repo, but this is simpler and made for…
I built a tiny LLM to demystify how language models workApr 2026 · github.com · ▲915Built a ~9M param LLM from scratch to understand how they actually work. Vanilla transformer, 60K synthetic conversations, ~130 lines of PyTorch. Trains in 5 min on a free Colab T4. The fish thinks the meaning of life is food. Fork it and swap the personality for your own character.
Qwen3.5 SmallMar 2026 · huggingface.co · ▲3550.8B-9B native multimodal w/ more intelligence, less compute
More ai this month
the category →
Paintings by AI models: programs run through a simulation of oil paint.
AI · 21d ago · stillwet.art
PixVerse R2▲381A real-time world model you can explore and change
AI · 30d ago · world.pixverse.video
Source: Show HN post ↗ · launch text quoted in part
Launched alongside, March 2025
the whole month →
Mimic Human Research & Save Findings in AI Knowledge Base
AI · 2025 · sider.ai
