MONDAY 7 SEPTEMBER 2026latent·wire21 PIECES ON FILE
← ModelsModels

OpenBMB releases MiniCPM5-2B, tops open models under 4B on AI index

OpenBMB released MiniCPM5-2B, a small open-weights language model that scores 15 on the Artificial Analysis Intelligence Index v4.2, the highest of any open-weights model at 4 billion parameters or below, according to a Reddit announcement in r/LocalLLaMA. The model is available on Hugging Face at openbmb/MiniCPM5-2B and on GitHub under the OpenBMB/MiniCPM repository. The release targets the small-model tier, where the benchmark result positions it ahead of comparable open models in its size class. The announcement did not detail the model's architecture, training data, license terms, or hardware requirements, and the underlying benchmark methodology was not described in the post. Details beyond the release and the index score remain limited at this stage.

Why it matters

A 2B-parameter open model topping the intelligence index in its size class could make capable local inference more accessible on modest hardware.