MBZUAI's IFM releases open PAN 2 world model

Nov 11, 2026
PAN 2 is a 29 billion parameter world model that turns text actions and optional images into simulated video. Its weights, training code and inference code are available under Apache 2.0.

The Institute of Foundation Models (IFM) at MBZUAI released PAN 2, a 29 billion parameter world model, as detailed on its Hugging Face model page. The model accepts a text action and an optional starting image, then generates up to 20 seconds of video at 704 by 1248 resolution and 24 frames per second.

PAN 2 combines conceptual reasoning from a language model with perceptual reasoning from a diffusion transformer. The language model converts an observation and requested actions into a structured description, while the diffusion model generates video showing the predicted result. The weights, training code and inference code are available under the Apache 2.0 license.

In IFM's own tests, as summarized by MarkTechPost, PAN 2 completed every step in 34% of 100 multi step tasks and completed 75.1% of individual steps. Removing the language model reasoning stage reduced full plan completion to zero. Each inference worker loads the full 29 billion parameter model in BF16, with block offloading available as a fallback.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Robotics Brief or Daily AI Brief.

Also, consider following us on social media:

AI & Robotics AI Brief AI Brief (X)
Free newsletter

Robotics Brief

Weekly coverage of AI-driven robotics advances in industrial automation, autonomous vehicles, and robotic systems.

Guide
Humanoid Robot Makers

Humanoid Robot Makers

Companies building humanoid robots, from startups to Tesla and Boston Dynamics

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more

You may also like

Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.