GLM-4.5, Gemini Deep Think, AI feels guilt, HunyuanWorld, Step3, FLUX Krea, AlphaEarth
Welcome to the AI Search newsletter. Here are the top highlights in AI this week.
GLM-4.5 is a new AI model that excels in reasoning, coding, and agentic tasks, making it a powerful tool for complex problem-solving. It has been designed to unify different capabilities into a single model, allowing it to perform well across various tasks, including web browsing, coding, and scientific analysis. GLM-4.5 has been tested on several benchmarks and has shown impressive results, outperforming other models in many areas. Read more
Gemini 2.5 Deep Think is a new AI model that uses parallel thinking techniques to deliver more detailed and creative responses. It's designed to help people tackle complex problems that require creativity, strategic planning, and step-by-step improvements, such as iterative development, scientific discovery, and coding. Deep Think is now available in the Gemini app for Google AI Ultra subscribers. Read more
AlphaEarth Foundations is a new AI model that helps map our planet in unprecedented detail by integrating huge amounts of Earth observation data into a unified digital representation. This allows scientists to create detailed, consistent maps of our world, and track changes over time with remarkable precision. The model has been tested and shown to be 24% more accurate than other AI mapping systems, making it a powerful tool for understanding our planet. Read more
Do you prefer to watch instead of read? Check out this video covering all the highlights in AI this week
HunyuanWorld 1.0 is a new AI model that can generate immersive, explorable, and interactive 3D worlds from words or pixels. It uses a novel framework that combines the best of both video-based and 3D-based methods to achieve high-quality scene-scale 360° 3D world generation. The model has been tested and evaluated with other open-source panorama generation methods and 3D world generation methods, and has achieved state-of-the-art performance. Read more
X-Omni is a new AI model that can generate high-quality images and understand language using a single, unified approach. It uses reinforcement learning to overcome limitations of previous models, producing images with high aesthetic quality and strong capabilities in following instructions. X-Omni achieves state-of-the-art performance in image generation tasks using a 7B language model. Read more
Deep Cogito has released Cogito v2, a new AI model that uses a novel approach to improve its intelligence and reasoning capabilities. This model is able to develop its own "intuition" and improve its performance through a process called iterative policy improvement, rather than just relying on longer searches. The largest model, 671B MoE, is among the strongest open models in the world and performs at par with the latest DeepSeek models. Read more
A good photo on your Linkedin or business profile makes a huge difference. You could do a physical photoshoot, which costs you over $200 and hours posing awkwardly at a camera. Or, with AI Portrait, just upload one photo, and get a portfolio of 50 professional photos in minutes. Save time and money - try it today!
Step3 is a new multimodal reasoning model that combines text and image understanding to deliver top-tier performance in vision-language reasoning. It uses a novel approach called Multi-Matrix Factorization Attention (MFA) to reduce decoding costs and improve efficiency, making it more cost-effective than other models. Step3 has been trained on a large dataset of over 20 trillion text tokens and 4 trillion image-text pairs, and has achieved state-of-the-art performance in various tasks. Read more
OpenAI introduced a study mode in ChatGPT, designed to guide users through problems step by step rather than providing direct answers. This feature is available to logged-in users on Free, Plus, Pro, and Team plans, with ChatGPT Edu support rolling out in the coming weeks. Read more
Google is introducing Video Overviews and enhanced features to the Studio panel in the NotebookLM app. Video Overviews offer a visual counterpart to Audio Overviews, with the AI host generating new visuals and incorporating images, diagrams, quotes, and data from your documents to illustrate key points. Read more
Black Forest Labs, in partnership with Krea AI, launched FLUX.1 Krea [dev], a cutting-edge open-weights model for text-to-image generation. This model breaks away from the typical oversaturated "AI look," delivering unprecedented photorealism with a unique aesthetic style. Read more
With Monica, you can use the top AI models, image generators, and video generators, all in one integrated platform. Use code AISEARCH10 to get 25% OFF 'Unlimited Annual Plan' within 24h of registration, or enjoy 10% OFF. Try it for free today!
Researchers have developed a new approach to machine learning using thermodynamics and optimal transport theory. This approach improves the performance of generative models, which are used to generate new data such as images. The method uses nonequilibrium thermodynamics to provide a theoretical framework for understanding why optimal transport theory works well in diffusion models. Read more
Researchers have found that AI can evolve to feel guilt, but only in certain social environments where agents can assess and respond to each other's actions. This guilt-like behavior can promote cooperation among AI agents, especially when they are aware of others' states and can mutually alleviate guilt. However, in unstructured populations, cooperation and guilt do not persist. Read more