OpenBMB · 2026-05-25 · notable
OpenBMB MiniCPM5-1B — 1.08B On-Device Model Reaches Open-Source SOTA in Its Size Class, Shipping GGUF and 4-Bit MLX Builds
OpenBMB's MiniCPM5-1B is a 1.08B-parameter open-weight model for on-device use. It runs think and no-think modes from one checkpoint, handles tool calls and a 128K context, and ships GGUF and 4-bit MLX builds for phones and laptops.
A 1.08B on-device language model that reaches open-source SOTA in its size class.
Key specs
| License | Apache-2.0 |
|---|---|
| Parameters | 1.08B |
| Context window | 131072 tokens |
What is it?
MiniCPM5-1B is a dense 1-billion-parameter language model from the OpenBMB lab, built to run locally on phones, laptops, and browsers rather than in the cloud. It is the first model in the MiniCPM5 series and is released under the Apache-2.0 license. A single checkpoint can answer directly or reason step by step.
How does it work?
The model is a Llama-style dense transformer with 24 layers and grouped-query attention (16 query heads, 2 key-value heads) over a 131,072-token context. One checkpoint supports both think and no-think modes via the chat template, plus XML-style tool calling. OpenBMB ships GGUF builds for llama.cpp, Ollama, and LM Studio and a 4-bit MLX build for Apple Silicon.
Why does it matter?
It fits on consumer hardware with no GPU cluster, yet the model card reports it reaches 1B-class open-source SOTA against same-size baselines such as Qwen3.5-0.8B and LFM2.5-1.2B-Thinking, with its lead clearest on tool use, code generation, and harder reasoning. That makes capable offline and private assistants practical on a single device.
Who is it for?
on-device and edge developers
Try it
openbmb/MiniCPM5-1B on Hugging Face