MBZUAI's IFM Releases PAN 2: An Open 29B World Model for Physical AI
The Institute of Foundation Models (IFM) at MBZUAI has released PAN 2, an open world model for physical intelligence. Given a scene and a text-specified action, PAN 2 simulates the likely outcome as video. IFM announced the release on X on November 10, 2026. Weights, training code, and inference code ship under the Apache 2.0 license. For robotics and agent developers, the pitch is straightforward: test a plan in simulation before acting on it.
TL;DR
- Size: A 29B-parameter dense diffusion transformer (MMDiT). Inputs are text plus an optional first frame. Output is 704×1248 video at 24 fps, up to 20 seconds.
- Runs on: Data-center GPUs. Each worker holds the full 29B model in BF16, with block offloading as a fallback.
- Performance: Leads 6 tested models on full plans. Trails MiniMax-H3 on single actions and human preference.
- Best: 34.0% full-plan completion on 100 multi-step tasks—the highest among all models tested (self-reported).
- Worst: About 2 in 3 plans miss a step. On the Physics-IQ system default, PAN 2 scores 37.48 versus 42.91 for MiniMax-H3.
- Bottom line: A fully open world model that leads on multi-step plans—but long plans still fail often, and hardware requirements are high.
What Is PAN 2?
PAN 2 is a general-purpose world model from IFM that simulates how a scene changes after an action. It follows the Generative Latent Prediction (GLP) architecture described in IFM's Critique of World Models paper and succeeds PAN, released in November 2025.
The design goal is simulative reasoning: an agent proposes plans, PAN 2 simulates each one, and the agent then acts on the best result.
via MarkTechPost