Suno 4.5 released

Suno 4.5 released

Imagine typing just two lines of emotion—and seconds later, an eight-minute track with velvety vocals floods your room. Suno now takes this dream to a new level with version 4.5: The model composes more dynamically, hits genres more precisely, and even mixes unusual pairings such as “Gregorian chant house” with surprising clarity.

V4.5 understands moods and instrument descriptions noticeably better, gives vocals more emotional depth, and weaves in subtle sound textures that previously only professional producers could achieve. A fresh prompt enhancement helper transforms rough ideas into detailed stage directions — ideal for anyone who wasn't born a sound engineer.

For the AI community, this means faster experimentation thanks to shorter generation times, realistic audio datasets without a studio budget, and finally room for narrative sound projects, because songs can now be up to eight minutes long and still sound consistent.

Where will this evolution lead? Perhaps to the first fully automatically produced LP – or to live jams in which humans and machines improvise together. What soundscapes would you create with v4.5?

Why it matters: Suno 4.5 dramatically lowers the threshold between an idea and a professional song. This could make music production as accessible as posting a tweet in the future – while also providing high-quality material for training future creative AIs.

The most important graph of our recent times: more compute in inference results in better performance

Humanoid robots in factories

Wired reports that humanoid robots are poised for commercial breakthrough in 2025: Boston Dynamics' electric Atlas is set to start work in Hyundai factories this year, while Agility, Figure & Co. will take on flexible material and logistics tasks. Thanks to advanced AI models such as Gemini Robotics, Goldman Sachs predicts that a $38 billion market will emerge. The combination of human-like mobility and generative AI promises faster changeovers, lower costs, and entirely new production processes.

Gemini Robotics – AI goes physical

Google DeepMind introduces Gemini Robotics: a vision-language-action model based on Gemini 2.0 that directly controls robots. It generalizes to unknown objects, replans tasks when changes occur, and increases the success rate of comparable systems by two to three times. With embodied reasoning and on-the-fly code generation, the same AI could range from laboratory arms to humanoid assistants, paving the way for truly universal service robots.

AI assistance in the operating room

The Orlando VA is the first veterans' hospital to test Moon Surgical's compact Maestro robot, which autonomously holds a camera and instruments, replacing up to two assistants. Twenty procedures since January have demonstrated how AI-supported camera tracking (“ScoPilot”) increases precision and efficiency while reducing the workload for staff. If the system proves successful, it could scale minimally invasive surgery cost-effectively—a game changer for overburdened hospitals.

China Creates Robot Powered by Living Human Brain Cells

Chinese researchers have built a robot controlled by a mini human brain grown from stem cells. This brain-on-chip system can move, dodge obstacles, and learn through experience, hinting at real cognitive function. By fusing biological neurons with digital circuits, scientists are stepping beyond AI into bio-digital intelligence. The breakthrough raises urgent ethical questions about consciousness, autonomy, and the line between machine and mind.