This week’s AI news highlights major advancements including OpenAI’s launch of ChatGPT apps, an apps SDK, and the Agent Kit for building AI workflows, alongside new models like the versatile Luminina Demu for image generation and the efficient Apriel 1.5B Thinker. Additionally, innovations span from AI-powered video tools and humanoid robots to Google’s free app builder Opal and Alibaba’s massive Ling 1 Trillion model, reflecting rapid progress in AI accessibility and integration.
This week in AI news has been packed with exciting developments across various domains. OpenAI held their DevDay event, unveiling several major updates including the introduction of apps within ChatGPT, allowing users to interact with third-party services like Canva, Spotify, and Zillow directly in the chat interface. They also launched an apps SDK for developers to build custom apps integrated with ChatGPT, potentially reaching over 800 million users. Additionally, OpenAI released an Agent Kit, a drag-and-drop workflow builder for AI agents, somewhat similar to tools like Zapier or n8n, but focused on the OpenAI ecosystem. Other notable releases include API access to GPT-5 Pro, their most advanced reasoning model, and Sora 2 Pro, a high-quality video generator with sound capabilities.
In the image generation and editing space, a new state-of-the-art model called Luminina Demu was introduced. This model excels at both generating images from text prompts and editing existing images with text instructions, similar to Nano Banana. It supports advanced features like style transfer, inpainting, outpainting, and control net capabilities such as edge maps and pose skeletons. The model is open-source and available on Hugging Face, with instructions for local deployment. Meanwhile, Nvidia released ChronoEdit, an AI image editor that applies edits by reasoning through temporal changes, making it useful for generating synthetic data for robotics and autonomous driving.
A remarkable tiny AI model named Apriel 1.5B Thinker was also announced, boasting impressive reasoning abilities despite having only 15 billion parameters. It outperforms many larger models on intelligence benchmarks and can run on most consumer-grade GPUs. This model was trained using a combination of diverse data and supervised fine-tuning without reinforcement learning, demonstrating that state-of-the-art performance can be achieved with efficient design and methodology. The model is open-source and available for local use, making it accessible for developers and researchers with limited resources.
Several innovative AI tools and robots were highlighted as well. Paper 2 Video is an AI that converts scientific papers into narrated presentation videos featuring a cloned voice and avatar of the presenter, streamlining academic presentations. Mimix allows mixing different fictional characters and styles in the same video, outperforming other video models in character preservation. In robotics, new humanoid robots like Deep Robotics’ waterproof DRR2 and Figure 03, designed for home environments, were showcased. These robots demonstrate capabilities such as object manipulation, household chores, and expressive walking gaits, hinting at a future where humanoid robots assist in daily life.
Finally, Google introduced Opal, a free drag-and-drop workflow builder for creating AI-powered apps, expanding access to 15 more countries. Opal enables users to build complex workflows integrating various AI models for tasks like blog post generation and video quizzes. Alibaba released Ling 1 Trillion, a massive AI model with one trillion parameters that outperforms many competitors in reasoning and coding benchmarks. Other notable mentions include Grock’s free video generator with audio, Samsung’s tiny but powerful 7-million-parameter recursion model excelling in reasoning tasks, and Google DeepMind’s Codemender, an AI agent that autonomously detects and fixes software vulnerabilities. Overall, this week’s AI advancements showcase rapid progress in making AI more accessible, efficient, and integrated into everyday applications.