We are on the verge of the biggest technological leap for AI. Not just because of intelligent super harnesses like Astra and the latest from frontier LLMs, but because of world models like Odyssey transforming how we model the physical world. Today, @odysseyml is launching Odyssey-3. Its ability to generalize across embodiments is something we could have only dreamed of just years ago. With a few hours of driving or flying video data, Odyssey-3 can help a car drive or a drone fly itself. With a few tens of hours of robotic or human data, Odyssey-3 can give robots guide paths for how to handle real world situations. In a matter of a few hours and recursive self learning, Odyssey-3 can beat humans in any game and even make games. Odyssey-3 can observe and understand the real world and generate video in real time with no latency. Odyssey-3 beats Odyssey-2 by 10X, while requiring 1000x less data, 100x lower training costs, and delivering 10x greater inference efficiency. In under a year, this company has gone from beating physics benchmarks to powering applications across a range of real-world environments. We’re already working with several leading game studios, robotics and autonomous technology companies, and interactive content producers that are seeing the superpowers of Odyssey in production. If you want to work with us, get in touch with @olivercameron , @jeffrey_hawke, or me.