MiniCPM5-2B

OpenBMB's dense 2B model for phones and laptops: 128K context, tool use and coding, Apache 2.0, with training data released.

In short. MiniCPM5-2B is a very small AI model that can run on a phone or an ordinary laptop. It is meant for local assistants, simple coding help and tool-using agents. It is released under Apache 2.0, and its developers also published much of the training data.

Summary

OpenBMB released MiniCPM5-2B on 7 September 2026, the second model of the MiniCPM5 series after MiniCPM5-1B in May. It is a dense transformer with about 2.5 billion parameters in total (about 2 billion excluding the embedding table), 42 layers and a context length of 131,072 tokens. It uses the standard Llama model architecture, which makes it easy to run in common tools.

OpenBMB designed it for on-device and resource-constrained use: local assistants, coding agents, tool-use workflows and reasoning. In its own comparison against Qwen3.5-2B, Gemma-4-E2B and LFM2.5-2.6B, plus several 3-to-4-billion-parameter models, MiniCPM5-2B scores an average of 53.9, ahead of the highest larger model at 51.1. These are the developer's own results on its chosen benchmarks, so check them against your own tasks.

Alongside the model, OpenBMB released the datasets behind it as part of its UltraData family, including a web pre-training set, a tiered code dataset and 500,000 agent training samples. The weights are about 5 GB in BF16, and GGUF, MLX, GPTQ and speculative-decoding variants are published. The license is Apache 2.0.

Good for

  • Local assistants and simple agents on a laptop, phone or small board
  • Experiments in fine-tuning, since the model, its base checkpoint and much of its training data are open
  • Cost-free commercial use under Apache 2.0
  • A small drafting or tool-calling model inside a larger system

Watch out for

  • Text only; for images use MiniCPM-V
  • At 2B parameters it will make more factual mistakes than larger models, so pair it with search or documents
  • The "beats 4B models" comparison is OpenBMB's own and covers a chosen set of benchmarks

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Estimates for the MiniCPM5-1B variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.

Variants Parameters BF16 FP8 / INT8 4-bit (Q4) Fits on
MiniCPM5-1B BF16 1.08B 2.48 GB 1.24 GB 0.745 GB Laptop or phone
MiniCPM5-2B BF16 2.52B 5.8 GB 2.9 GB 1.74 GB Laptop or phone

Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.

VRAM calculator

The license in plain words

One of the most widely used permissive licenses. You may use, change and sell products built on the model with only light conditions.

More about Apache 2.0

See also

Frequently asked questions

What is MiniCPM5-2B?
MiniCPM5-2B is a very small AI model that can run on a phone or an ordinary laptop. It is meant for local assistants, simple coding help and tool-using agents. It is released under Apache 2.0, and its developers also published much of the training data.
Can I use MiniCPM5-2B commercially?
Yes. MiniCPM5-2B is released under Apache 2.0, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
Is MiniCPM5-2B open source?
MiniCPM5-2B is released under Apache 2.0, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
Can I run MiniCPM5-2B on my own computer?
Yes. The smallest variant, MiniCPM5-1B, needs about 0.745 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
When was MiniCPM5-2B released?
MiniCPM5-2B was released by OpenBMB on September 7, 2026.
How long a context window does MiniCPM5-2B have?
MiniCPM5-2B supports a context window of up to 128K tokens.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 8, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
LFM3 Liquid AI 1.73B – 2.17B 256K LFM Open 1.0 (Conditional) 1.19 GB Laptop or phone
Gemma 4 Google DeepMind 5.1B – 30.7B 256K Apache 2.0 (Permissive) 3.52 GB Laptop or phone
Granite 4.2 IBM 3.7B – 29.3B 128K Apache 2.0 (Permissive) 2.55 GB Laptop or phone
Nanbeige4.2-3B Nanbeige LLM Lab 4.17B 256K Apache 2.0 (Permissive) 2.88 GB Laptop or phone