MiniCPM-V 4.6: Tsinghua Spinoff Open-Sources a 1.3B Multimodal Model That Runs on a Single RTX 4090

OpenBMB, in collaboration with Tsinghua University, has launched MiniCPM-V 4.6, a multimodal AI model with 1.3 billion parameters. This new model outperforms larger models such as Alibaba's Qwen3.5-0.8B and Google's Gemma4-E2B-it in vision and language tasks, offering significantly faster inference speeds. MiniCPM-V 4.6 features key architectural innovations, including an early-exit strategy in image processing that reduces subsequent computations while maintaining image representation quality.