GitHub radar
MiniCPM5 2B: On-device Agent Model
OpenBMB released MiniCPM5 2B — a 2.5B model with 131K token context and native tool calling, built for local deployment on any laptop.
OpenBMB — the Chinese AI lab behind the MiniCPM model family — released MiniCPM5 2B, the second model in the MiniCPM5 series. It is a 2.5B dense transformer built for on-device deployment: laptops, edge hardware, and resource-constrained scenarios where a full server GPU is not available. The model features a 131,072-token context window and native tool calling, enabling agent workflows without any cloud dependency. In benchmarks, it achieves an average score of 53.9, exceeding all compared models in its class. It runs locally via Ollama, LM Studio, llama.cpp, and MLX.
Why a vibe-coder should care
If you run AI agents through the cloud, MiniCPM5 2B gives you the same tool-calling and long-context capabilities on your own laptop — cheaper, without network latency, and without sending your data to a third party.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install MiniCPM5 2B locally via Ollama: https://huggingface.co/openbmb/MiniCPM5-2B — if it's not in Ollama yet, find a compatible GGUF build, run it and show me how to use it
Обычный ноутбук от 8 ГБ оперативки — в четырёхбитной версии модель занимает около 1.5 ГБ
Open on Hugging Face▌ More finds