Some links may be affiliate links. We may earn a small commission at no extra cost to you. Learn more
Open-source AI world model by Decart and Etched that generates real-time Minecraft-style interactive gameplay at 20 FPS using next-frame prediction, with no traditional game engine required.
Editor's take: “Spatial/world generation is an emerging and exciting capability” — Sohail Akhtar
Free tier (verified July 2026): Free browser demo, no signup
Reviewed by Sohail Akhtar
Lead Editor & Founder
What we like
Limitations
✓ Free tier re-verified July 2026: Free browser demo, no signup
| Plan | Details |
|---|---|
| Free | Full browser-based demo accessible at no cost with no account required. Open-source 500M parameter model weights and code available on GitHub for free local deployment under the project's open-source license. |
| Paid | No paid tier. Oasis is a research project with no commercial subscription or access fee. |
Oasis is free to access. The browser-based demo is publicly available at no cost with no registration required. The open-source 500-million-parameter model weights and code are freely available on GitHub for local deployment. Decart does not currently charge for access to the demo or the model. Local deployment requires hardware capable of running the model, and the hosted demo runs on Decart's NVIDIA H100 GPU infrastructure.
Quick Summary
Oasis is an AI-generated interactive world model developed jointly by Decart, an Israeli AI company, and Etched, a hardware startup, released publicly on October 31, 2024 as the first playable real-time AI-generated game environment operating without a traditional game engine. It uses next-frame prediction—a transformer-based architecture combining a Vision Transformer (ViT) encoder and a diffusion model—to generate physics, game rules, and visuals in real time at 20 frames per second, taking player keyboard and mouse inputs and predicting what the game world looks like after each action. Oasis was trained on millions of hours of Minecraft gameplay footage and is freely accessible via a browser demo and as an open-source 500-million-parameter model on GitHub.
Associated Tags
AI world model, real-time AI generation, next-frame prediction, AI game engine, open source game AI, generative game world, Minecraft AI model, interactive AI simulation
Who should use Oasis AI Game?
Discover practical workflows and real-world scenarios where Oasis AI Game delivers key solutions.
An AI researcher uses the Oasis open-source codebase to study the next-frame prediction architecture and dynamic noising technique as a reference implementation for their own world model research.
A game developer explores the Custom Worlds feature by uploading concept art images to generate initial playable environments, evaluating how image-to-world generation could accelerate early-stage level prototyping.
A technical educator uses the Oasis browser demo during a presentation on AI capabilities to give a non-technical audience a direct, interactive demonstration of what real-time generative world models can currently produce.
A computer science student runs the open-source 500M model locally on a research GPU to experiment with fine-tuning the architecture on a different gameplay dataset for a class project on generative models.
A journalist covering AI gaming technology uses the public demo to personally experience and accurately describe the current capability and limitations of AI-generated interactive environments for a feature article.
An AI hardware researcher studies the Oasis architecture as a benchmark case for evaluating the performance advantages of Etched's Sohu Transformer ASIC over general-purpose GPU deployments for real-time world generation.
Skywork AI's 1.8B open-source interactive world model generating real-time 25 FPS gameplay from keyboard and mouse inputs, with long-sequence consistency and free weights on GitHub and Hugging Face.
Tencent's open-source AI model that generates interactive, action-controllable game video sequences from a single image and keyboard inputs.
World Labs' real-time generative world model that creates persistent, navigable 3D scenes from a single image on one H100 GPU.
Open-source AI model by Tencent that generates explorable, interactive 3D worlds from text or image inputs using panoramic scene reconstruction.
