OpenBMB Open Models: Model List, Licenses and Articles

September 26, 2026September 27, 2026

About OpenBMB

OpenBMB (Open Lab for Big Model Base) is a research organization that, on its openbmb organization on Hugging Face, “aims to build foundation models and systems towards AGI". It publishes 173 models.

OpenBMB focuses on small models that run on devices such as phones and laptops. Its main lines are the MiniCPM language models, MiniCPM-V, which also reads images, and VoxCPM for speech synthesis, all released in small sizes.

What It Releases

• MiniCPM5 (language models): MiniCPM5-1B from May 2026 and MiniCPM5-2B from September. The 1B model card claims “1B-class open-source SOTA", with its advantage most visible in agentic tool use, code generation and difficult reasoning. Thinking can be switched on and off (our family page has a detailed explanation of the 2B).
• It also releases weights from intermediate training stages: Base, Midtrain and SFT checkpoints for MiniCPM5.
• MiniCPM-V (models that also read images): the 4.6 Thinking version is small at about 1.3B, with official AWQ and bitsandbytes quantized builds.
• VoxCPM (speech synthesis): including VoxCPM2 (April 2026).
• BitCPM-CANN: 1.58-bit language models (0.5B to 8B) with ternary weights (−1, 0, 1), trained on Huawei Ascend NPUs. According to the model card, the 1B, 3B and 8B models retain 95.7% to 97.2% of full-precision performance while reducing inference memory about sixfold.

Licenses

The main models we checked, including MiniCPM5, MiniCPM-V 4.6, VoxCPM2 and BitCPM-CANN, are Apache-2.0. Some models (such as the research model JustRL) have no license field, so read each model’s license text before use. This summary is not legal advice.

Running Them Locally

• Official GGUF and MLX builds are plentiful. OpenBMB itself publishes GGUF, MLX (for Apple Silicon) and GPTQ builds of MiniCPM5 1B and 2B, which run directly in llama.cpp, Ollama and LM Studio. There are also GGUF builds of the DSpark helper models that speed up inference.
• The main BitCPM-CANN weights are “pseudo-quantized". According to the model card, the ternary values are stored in floating point, so they load like ordinary models. Use the GGUF builds to actually save memory.
• Most models are small enough to try easily, not only on consumer GPUs but on CPU-only laptops.

Sources: the openbmb organization card; model cards of openbmb/MiniCPM5-1B and openbmb/BitCPM-CANN-8B; the model list of the openbmb organization on Hugging Face (all as of 2026-09-26).

Our Coverage and Data

Local Model Watch has published 1 article(s) on OpenBMB’s own models, 0 on third-party fine-tunes and quantized builds based on them, and 0 from its official blog. The model list below is collected by code from Hugging Face every day. Part of our publisher index; model families link to their family pages.

At a Glance

Recent Models on Hugging Face

The publisher lists 173 models in total on Hugging Face. Below are up to 15 of the most recent generative models (text, image, video, audio and so on), the scope of this site, per organization. Classifiers, feature extractors, research components and repositories without a model card are omitted, and the publisher’s own quantized builds and helper variants are folded into the row of the original model. Parameter counts are left blank for repositories that pack weights into integer types, where Hugging Face’s count is not the real parameter count.

From the Hugging Face API. Last changed 2026-09-27 (JST). Collected by code.

Licenses of Its Own Models (as recorded in our articles)

The license can differ from model to model even within one publisher. This is not legal advice; check each model’s license text before use.

By Model Family

Articles on Its Own Models

Last updated 2026-09-27 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log and the Hugging Face API.