ModelBest

ModelBest builds a family of efficient AI models spanning multiple modalities. MiniCPM-O is a full-duplex omni model supporting streaming video, audio, text input, and speech output, while MiniCPM-V is an on-device vision model that runs on phones with just 6GB of memory across iOS, Android, and HarmonyOS. The company's embodied intelligence series includes a 1.5B-parameter VLA model that delivers performance comparable to 3-4B models with one-minute native memory and streaming inference, alongside a tracking model that supports fully offline visual tracking. ModelBest also offers an end-to-end tokenizer-free speech synthesis model covering 30 languages and 9 Chinese dialects, supporting voice design from text descriptions, zero-shot voice cloning, and 48kHz CD-quality output under the Apache 2.0 license.