OpenBMB Releases MiniCPM5-2B, Setting New Benchmark for Sub-4B Open-Weights Models
OpenBMB has officially released MiniCPM5-2B, a 2-billion-parameter open-weights language model. It scored 15 on the Artificial Analysis Intelligence Index v4.2, achieving the highest benchmark score of any open-weights model under 4B parameters. This release advances the capabilities of small language models (SLMs), making powerful AI capabilities accessible for local execution on consumer hardware and edge devices. High-performing sub-4B models significantly lower deployment costs and reduce reliance on cloud-hosted proprietary APIs. MiniCPM5-2B earned its top score on the Artificial Analysis benchmark, which evaluates models across agent tasks, coding, general capabilities, and scientific reasoning. The model weights and source code have been made publicly available on Hugging Face and GitHub.
## BACKGROUND
Small Language Models (SLMs) focus on optimizing efficiency and accuracy within low resource constraints compared to massive enterprise LLMs. The Artificial Analysis Intelligence Index is a widely recognized benchmark suite that evaluates and ranks models using a weighted average across key operational domains.