OpenBMB has released MiniCPM-o 2.6, an 8 billion parameter multimodal large language model (LLM) that reportedly outperforms competitors such as GPT-4o, Gemini 1.5 Pro, and Sonnet in various tasks. The model achieves an average score of 70.2 on OpenCompass for visual tasks, while also excelling in bilingual real-time speech recognition and transcription, utilizing 75% fewer vision tokens. MiniCPM-o 2.6 supports over 30 languages and is capable of real-time video and audio understanding. It is designed to run on edge devices, including smartphones and tablets. Additionally, fine-tuning options for the model have been made available through LLaMA-Factory.