12 GB accelerator memory
Q2_0_g128 GGUF · reduced context · llama.cpp / MLX
The model card reports roughly 8.4–8.7 GB at 4K–10K context; 12 GB is the safer reduced-context boundary.
Prism ML
Text model
Ternary Bonsai 27B
A 27B multimodal reasoning model compressed into a laptop-sized ternary build.
ChatCoding
Capabilities
What you can use it for.
The most useful reasons to choose this model, without making you read through its repository first.
Ternary weightsVisionTool callingLong context
01
Local reasoning assistants
02
Tool-using workflows
03
Multimodal analysis
For providers
Want to serve this model?
The provider requirements live here, separate from the information you need to choose and use the model.
Starting pointCPU · 16 GB RAM
Model source
CPU + 16 GB RAM
Q2_0_g128 GGUF · reduced context · llama.cpp
The 7.2 GB packed model supports CPU inference; throughput depends heavily on memory bandwidth and context length.
16 GB accelerator memory
Q2_0_g128 GGUF + Q4 KV · llama.cpp / MLX
The published full-context peak is about 12.8 GB with a Q4 KV cache, so 16 GB leaves practical runtime headroom.
Source and license
Check the original model.
OpenMayhem links back to the source repository so you can inspect the model card, files, license, limitations, and creator guidance directly.
