Shanghai AI Lab's OpenGVLab has released InternVL3.5, an open-source multimodal AI model family that demonstrates state-of-the-art performance in image and text tasks. InternVL3.5 achieves 16% better reasoning performance and is 4.05 times faster in inference compared to its predecessor, InternVL3. The model incorporates a scalable reinforcement learning framework called Cascade Reinforcement Learning, which combines offline and online learning to enhance reasoning capabilities. The family includes models ranging from 1 billion to 38 billion parameters, with mixture-of-experts (MoE) variants scaling up to 241 billion parameters. InternVL3.5 is available under the Apache 2.0 license and has been integrated into the LocalAI platform, where versions with 4 billion, 8 billion, 14 billion, and 30 billion parameters are accessible for deployment. Meanwhile, Alibaba has updated its open-source video-generating AI model, maintaining a rapid development pace to compete with other Chinese and US AI advancements. Additionally, OpenBMB released MiniCPM-V 4.5, a new multimodal large language model optimized for image, multi-image, and video understanding, capable of running on mobile devices and featuring 96 times video token compression for high-frame-rate and long video reasoning.