ModelBest
A Tsinghua University-affiliated spinoff commercializing the MiniCPM family of efficient on-device foundation models. MiniCPM models are engineered for smartphones, PCs, automotive systems, and edge devices with GPT-4V-level multimodal performance at 1-9B parameters. Operates OpenBMB, an open-source research lab for large model development.
MiniCPM-V 2.5 (August 2024) outperformed GPT-4V on vision benchmarks while requiring <3B parameters. Latest MiniCPM5-1B (May 2026) achieves 1B-class SOTA with 42.57 average benchmark score. Strong focus on on-device efficiency and Apache 2.0 open-weight releases.
No products catalogued yet. Press Refresh above to ask the research agent.
No people linked.
No citations stored yet. Run the research agent to attach per-claim sources.