Moonlake AI is developing foundation models for creating real-time interactive content, including simulations and games, directly from natural language. Where most generative AI produces static text, images, or video, Moonlake's models are designed to output playable, interactive worlds, letting a creator describe an experience and watch a working simulation come to life. The company frames this as "vibe coding" interactive worlds, lowering the barrier so that people without engineering or game-development backgrounds can build dynamic experiences in minutes.
Technically, Moonlake is building multimodal reasoning models that generate entire interactive simulations and games from prompts. This requires the system to reason about state, physics, rules, and player interaction rather than simply rendering an image, positioning Moonlake squarely in the emerging world-models category alongside efforts to make AI that understands and simulates dynamic environments. The team has framed the work as reasoning models for interactivity, distinct from passive media generation.
Moonlake was founded by Fan-Yun Sun and Sharon Lee while working out of the Stanford AI Lab. Sun earned a PhD in computer science at Stanford and previously worked at NVIDIA on large-scale 3D world generation for AI training and foundation models, experience that directly informs Moonlake's approach to building interactive world models.
The company emerged from stealth in 2025 with $28 million in seed funding co-led by Threshold Ventures, AIX Ventures, and NVIDIA's venture arm NVentures, alongside more than a dozen prominent AI researchers and founders. Angel backers reportedly include Steve Chen, Jeff Dean, Naval Ravikant, Ian Goodfellow, and Guillermo Rauch, among others from OpenAI, DeepMind, Hugging Face, and Stability AI. After surpassing 10,000 users on its waitlist, Moonlake opened beta access to its world-modeling agent.