Google has unveiled Gemma3n, a lightweight multimodal AI model capable of running on devices with just 2GB RAM. The model supports text, image, video, and audio processing without cloud connectivity, marking a significant advancement in edge AI technology.
ByteDance has open-sourced BAGEL, a cutting-edge multimodal AI model with 7B active parameters. The model excels in text-image generation, editing, and outperforms competitors like Qwen2.5-VL and SD3. Its unique Mixture of Transformers architecture enables advanced capabilities from frame prediction to 3D operations.
ByteDance has unveiled Seed1.5-VL, a powerful 20B-parameter multimodal AI model that achieves state-of-the-art performance in 38 tasks. The model excels in visual-language understanding, video processing, and logical reasoning while offering cost-effective API access through Volcano Engine.
Google's Gemma AI model has surpassed 150 million downloads, highlighting its popularity among developers. The model's multimodal capabilities and multilingual support set it apart, though licensing concerns persist.
ByteDance's Seed team unveiled its multimodal AI model, Seed1.5-VL, positioning it as a strong competitor to Google's Gemini 2.5 Pro. The model excels in visual reasoning and agent tasks while reducing computational costs, though challenges remain in fine-grained perception.
© 2024 - 2025 Summer Origin Tech
