← CanIRunLocal · All hardware · All models

Can I run Bonsai 2 4B (ternary) locally?

Prism ML · 4B parameters · chat, vision, reasoning.

Ollama weight files: Q2_0 1.6 GB (bonsai-2:4b).

Yes on 56 machines in this list. The smallest is M1 8GB: Q2_0, about 2.8 GB, roughly 34 tokens per second, via Ollama. ollama run bonsai-2:4b

Setup on DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte)

Open Bonsai 2 4B (ternary) in the checker

Hardware that fits

HardwareRuntimeQuantMemoryFitSpeedCommand
M1 8GBOllamaQ2_02.8 GBfits in fast memory~34 tok/sollama run bonsai-2:4b
M2 8GBOllamaQ2_02.8 GBfits in fast memory~50 tok/sollama run bonsai-2:4b
M3 8GBOllamaQ2_02.8 GBfits in fast memory~50 tok/sollama run bonsai-2:4b
RTX 3060 Ti 8GBOllamaQ2_02.8 GBfits in fast memory~168 tok/sollama run bonsai-2:4b
RTX 3070 8GBOllamaQ2_02.8 GBfits in fast memory~168 tok/sollama run bonsai-2:4b
RTX 4060 8GBOllamaQ2_02.8 GBfits in fast memory~102 tok/sollama run bonsai-2:4b
RX 7600 8GBOllamaQ2_02.8 GBfits in fast memory~108 tok/sollama run bonsai-2:4b
RTX 3080 10GBOllamaQ2_02.8 GBfits in fast memory~285 tok/sollama run bonsai-2:4b
M1 16GBOllamaQ2_02.8 GBfits in fast memory~34 tok/sollama run bonsai-2:4b
M2 16GBOllamaQ2_02.8 GBfits in fast memory~50 tok/sollama run bonsai-2:4b
M3 16GBOllamaQ2_02.8 GBfits in fast memory~50 tok/sollama run bonsai-2:4b
M4 16GBOllamaQ2_02.8 GBfits in fast memory~60 tok/sollama run bonsai-2:4b
RTX 3060 12GBOllamaQ2_02.8 GBfits in fast memory~135 tok/sollama run bonsai-2:4b
RTX 4070 12GBOllamaQ2_02.8 GBfits in fast memory~189 tok/sollama run bonsai-2:4b
RTX 5070 12GBOllamaQ2_02.8 GBfits in fast memory~252 tok/sollama run bonsai-2:4b
RTX 4060 Ti 16GBOllamaQ2_02.8 GBfits in fast memory~108 tok/sollama run bonsai-2:4b
RTX 4070 Ti Super 16GBOllamaQ2_02.8 GBfits in fast memory~252 tok/sollama run bonsai-2:4b
RTX 4080 16GBOllamaQ2_02.8 GBfits in fast memory~269 tok/sollama run bonsai-2:4b
RTX 5080 16GBOllamaQ2_02.8 GBfits in fast memory~360 tok/sollama run bonsai-2:4b
RX 7800 XT 16GBOllamaQ2_02.8 GBfits in fast memory~234 tok/sollama run bonsai-2:4b
RX 9070 XT 16GBOllamaQ2_02.8 GBfits in fast memory~240 tok/sollama run bonsai-2:4b
M2 24GBOllamaQ2_02.8 GBfits in fast memory~50 tok/sollama run bonsai-2:4b
M3 24GBOllamaQ2_02.8 GBfits in fast memory~50 tok/sollama run bonsai-2:4b
M4 24GBOllamaQ2_02.8 GBfits in fast memory~60 tok/sollama run bonsai-2:4b
RTX 3090 24GBOllamaQ2_02.8 GBfits in fast memory~351 tok/sollama run bonsai-2:4b
RTX 4090 24GBOllamaQ2_02.8 GBfits in fast memory~378 tok/sollama run bonsai-2:4b
RX 7900 XTX 24GBOllamaQ2_02.8 GBfits in fast memory~360 tok/sollama run bonsai-2:4b
M1 Pro 32GBOllamaQ2_02.8 GBfits in fast memory~100 tok/sollama run bonsai-2:4b
M2 Pro 32GBOllamaQ2_02.8 GBfits in fast memory~100 tok/sollama run bonsai-2:4b
M4 32GBOllamaQ2_02.8 GBfits in fast memory~60 tok/sollama run bonsai-2:4b
M3 Pro 36GBOllamaQ2_02.8 GBfits in fast memory~75 tok/sollama run bonsai-2:4b
Mac Studio M4 Max (32c GPU) 36GBOllamaQ2_02.8 GBfits in fast memory~205 tok/sollama run bonsai-2:4b
Mac Studio M5 Max (32c GPU) 36GBOllamaQ2_02.8 GBfits in fast memory~230 tok/sollama run bonsai-2:4b
RTX 5090 32GBOllamaQ2_02.8 GBfits in fast memory~672 tok/sollama run bonsai-2:4b
M4 Pro 48GBOllamaQ2_02.8 GBfits in fast memory~137 tok/sollama run bonsai-2:4b
Mac Studio M4 Max (40c GPU) 48GBOllamaQ2_02.8 GBfits in fast memory~273 tok/sollama run bonsai-2:4b
Mac Studio M5 Max (40c GPU) 48GBOllamaQ2_02.8 GBfits in fast memory~307 tok/sollama run bonsai-2:4b
M1 Max 64GBOllamaQ2_02.8 GBfits in fast memory~200 tok/sollama run bonsai-2:4b
M2 Max 64GBOllamaQ2_02.8 GBfits in fast memory~200 tok/sollama run bonsai-2:4b
M3 Max 64GBOllamaQ2_02.8 GBfits in fast memory~200 tok/sollama run bonsai-2:4b
M4 Max 64GBOllamaQ2_02.8 GBfits in fast memory~273 tok/sollama run bonsai-2:4b
Mac Studio M4 Max (40c GPU) 64GBOllamaQ2_02.8 GBfits in fast memory~273 tok/sollama run bonsai-2:4b
Mac Studio M5 Max (40c GPU) 64GBOllamaQ2_02.8 GBfits in fast memory~307 tok/sollama run bonsai-2:4b
DGX Spark 64GB (ASUS, Dell, HP, MSI, Acer, Gigabyte)OllamaQ2_02.8 GBfits in fast memory~128 tok/sollama run bonsai-2:4b
Mac Studio M3 Ultra 96GBOllamaQ2_02.8 GBfits in fast memory~410 tok/sollama run bonsai-2:4b
Mac Studio M5 Ultra 96GBOllamaQ2_02.8 GBfits in fast memory~600 tok/sollama run bonsai-2:4b
M3 Max 128GBOllamaQ2_02.8 GBfits in fast memory~200 tok/sollama run bonsai-2:4b
M4 Max 128GBOllamaQ2_02.8 GBfits in fast memory~273 tok/sollama run bonsai-2:4b
Mac Studio M4 Max (40c GPU) 128GBOllamaQ2_02.8 GBfits in fast memory~273 tok/sollama run bonsai-2:4b
Mac Studio M5 Max (40c GPU) 128GBOllamaQ2_02.8 GBfits in fast memory~307 tok/sollama run bonsai-2:4b
DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte)OllamaQ2_02.8 GBfits in fast memory~128 tok/sollama run bonsai-2:4b
M2 Ultra 192GBOllamaQ2_02.8 GBfits in fast memory~400 tok/sollama run bonsai-2:4b
Mac Studio M3 Ultra 256GBOllamaQ2_02.8 GBfits in fast memory~410 tok/sollama run bonsai-2:4b
Mac Studio M5 Ultra (64c GPU) 256GBOllamaQ2_02.8 GBfits in fast memory~600 tok/sollama run bonsai-2:4b
Mac Studio M5 Ultra (80c GPU) 256GBOllamaQ2_02.8 GBfits in fast memory~600 tok/sollama run bonsai-2:4b
Mac Studio M3 Ultra (80c GPU) 512GBOllamaQ2_02.8 GBfits in fast memory~410 tok/sollama run bonsai-2:4b

Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-18.

A project by OnDevice Labs · llms.txt