Measured model
Measured on a Mac Studio (M5 Max, 36GB)MiniCPM5-2B
OpenBMB · Ollama name hf.co/openbmb/MiniCPM5-2B-GGUF:Q4_K_M
On a Mac Studio (M5 Max, 36GB), MiniCPM5-2B answers a simple question in about 1.5 s, gets 73% of the Japanese knowledge questions right, 90% of the math problems, and uses 2GB of memory.
Results
| Measure | Result | Among measured models | Note |
|---|---|---|---|
| Reply time | about 1.5 s | 7 of 33 | Until one simple common-sense question is answered (median of 50) |
| Knowledge (high school to university) | 73% | 22 of 33 | 200 questions, ±7 |
| Math (word problems) | 90% | 22 of 33 | 100 questions, ±7, 1 cut off (counted wrong) |
| Memory used | 2GB | 4 of 33 | |
| EN→JA translation | 23 pts | 29 of 30 | out of 100 (chrF), about 9 s a sentence |
| JA→EN translation | 48 pts | 28 of 30 | out of 100 (chrF), about 8.3 s a sentence |
"Among measured models" is the place among the 33 models measured on this site (lower is better for time and memory).
Measured on other Macs
The numbers and places on this page come from the Mac above alone.
Try it on your Mac
- Install the Mac version of Ollama from ollama.com and start it.
- Type this in Terminal. The first time, the model downloads; then you can chat with it.
ollama run hf.co/openbmb/MiniCPM5-2B-GGUF:Q4_K_M
It used 2GB here. To use it comfortably alongside other apps, a Mac with 8GB of memory or more is a good guide (the model then takes about half the memory).
About the model
- Maker
- OpenBMB
- Size
- 2.52B parameters
- Quantization
- Q4_K_M
- Thinks before answering
- Yes
- Longest input at once
- 131,072 tokens
- Released (Hugging Face)
- Sep 6, 2026
- Generation speed
- 187.6 tok/s
- Start-up (loading)
- about 0.8 s
- Measured on
- Nov 2, 2026
- Ollama
- 0.34.3
- Hugging Face
- openbmb/MiniCPM5-2B
- Series
- MiniCPM
What the parts of a name mean (size, quantization and so on): how to read a model name.
Models of similar memory
- LFM2.5-2.6B (2GB of memory, knowledge 73%, reply about 2.3 s)
- DeepSeek-R1-Distill-Qwen-1.5B (1.5GB of memory, knowledge 23%, reply about 1.7 s)
- LFM2.5-1.2B-JP-202606 (1GB of memory, knowledge 51%, reply about 0.2 s)
- NVIDIA-Nemotron-3-Nano-4B-BF16 (3.3GB of memory, knowledge 72%, reply about 1.9 s)
- Qwen3.5-4B (3.5GB of memory, knowledge 81%, reply about 29.7 s)
Compare the main models on the text page. The series page lists every model measured.
Task recommendations and capabilities · Measurements by model