MongoDB - Build AI That Scales
Mini toolsMatrix Game 3.0
By Skywork AI
Matrix-Game 3.0 is an open-source memory-augmented interactive world model designed for action-conditioned video generation rather than standard offline text-to-video. The model card describes a three-stage system: a large-scale data engine producing Video-Pose-Action-Prompt training data, a memory-augmented Diffusion Transformer with an error buffer for long-horizon consistency, and an optimized deployment stack using few-step sampling, INT8 quantization, and distillation. Skywork says it achieves 720p at 40 FPS in real time with a 5B model, supports minute-long stable generation, and also scales to a 28B MoE variant for higher quality and generalization.
New
Multimodal
Gen 3
Overview
Matrix-Game 3.0 is Skywork’s open-source interactive world model for real-time long-form video generation. It is built to generate 720p streaming video conditioned on images, prompts, and actions, with long-horizon memory for consistency over extended sequences, and is optimized for interactive use with few-step sampling, quantization, and distillation.
About Skywork AI
View Company ProfileTools using Matrix Game 3.0
No tools found for this model yet.
Last updated: April 1, 2026
