According to MarkTechPost, PAN 2 couples a large language model planner with a 29-billion-parameter diffusion transformer. The publication describes the system as a world model intended for physical AI and released under the Apache 2.0 license.
MarkTechPost reports that PAN 2 can simulate up to 20 seconds of action outcomes as video, linking language-based planning with a visual representation of possible physical actions.
The supplied record leaves the release chronology unclear: its URL is dated November 10, 2026, while the publication timestamp is November 11, 2026. The release itself is reported by MarkTechPost, and the supplied evidence does not independently establish a release date or location.