Agentic WeeklyGet The Weekly Verdict, freeSubscribe free

Reported News

MBZUAI's IFM releases PAN 2, an open 29B world model that simulates actions as video

The Institute of Foundation Models (IFM) at MBZUAI released PAN 2 on Nov 10, 2026, an Apache 2.0 world model whose 29B-parameter diffusion backbone turns an observation and text actions into "a video of up to 20 seconds," per its model card. IFM says it ranks first among open-weight models on the Physics-IQ Verified leaderboard.

By Agentic Weekly · Evidence: Reported · Editorial standards

Share on X
Changes reported
5
listed below, each from a source we read
We tested
Not tested yet
reported from sources; we did not run it ourselves
Gaps we flag
4
things we could not verify
Sources
4
3 primary, linked below

What changed

Changes the sources report, with the evidence status of each
#What the source saysStatus
01PAN 2 is a general-purpose world model: given a current observation (an optional image plus text) and actions described in text, it renders the predicted outcome as video, 704x1248 at 24 fps and up to 20 seconds long, according to the model card.Reported
02The system splits the work in two. An LLM (the conceptual reasoner) writes a structured, time-stamped description of the next world state, and a 29B-parameter MMDiT diffusion model trained from scratch turns that into video latents, decoded by the Wan 2.2 VAE with Qwen3.5-9B as the frozen text encoder.Reported
03IFM is releasing the diffusion weights plus full training and inference code under Apache 2.0; the model card says PAN 2 also runs in Diffusers and vLLM, and ships a prompt-rewriter agent skill for Codex and Claude Code that enforces a 1,536-token caption limit.Reported
04On IFM's own evaluations, the press release says PAN 2 placed just behind MiniMax-H3 and ahead of four other models across six simulation dimensions, completed the most full multi-step plans of any model tested, and ranks third overall and first among open-weight models on the Physics-IQ Verified leaderboard.Reported
05IFM recommends running PAN 2 behind a separate pan2-guardrail package; in internal tests it cut the share of harmful text-to-video requests that produced an unsafe video from 33.9% to 1.5%, while also blocking 11.5% of benign text-to-video requests.Reported

reported by the sourcemeasured or run by usnot verified

Primary source: Hugging Face: IFM/PAN2 model card. We are the fast, independent index; the vendor’s own page is the source of record for this release.

In the source’s words

“Given a current observation (an optional image plus text) and actions described in text, it simulates the outcome as a video of up to 20 seconds.”

What we have not verified

  • We did not run PAN 2; every benchmark, ranking and guardrail figure comes from IFM's model card and press release.
  • The model card does not state GPU memory requirements or generation time per clip; its inference script loads the full 29B diffusion model on every GPU worker.
  • The model card lists Claude Opus 5.5 as the conceptual reasoner for the public leaderboard entry, so the Physics-IQ result reflects PAN 2 paired with a closed LLM, not the diffusion model alone.
  • We could not open IFM's technical blog post directly (it returned an error to our fetcher), so details that appear only there are not reflected here.

This brief is Reported: written from the sources below, not from running the release ourselves.

Sources

Spotted an error? See our corrections policy or email the address on the contact page. Markdown version for agents: /news/ifm-pan-2-open-world-model.md

Want these briefs in your Google Top Stories? Add Agentic Weekly as a preferred source on Google.

Newsletter

The Weekly Verdict. Free, every Monday.

A free email every Monday: the week's few big AI releases, then 5 short use cases for the new models, each with a prompt you can paste.

Free. Double opt-in. One click to unsubscribe.