PAN 2 vs Gemma 4 12B: Een simulator en een redeneerder zijn geen rivalen

Every comparison page on the internet wants a winner, and this one cannot honestly have one, because PAN 2 and G​emma 4 12B do not perform the same task. PAN 2 is a 29-billion-parameter video diffusion transformer from MBZUAI's Institute of Foundation Models that takes an image and a description of actions and returns up to twenty seconds of 704×1248 video showing what happens next. G​emma 4 12B Unified is an 11.95-billion-parameter encoder-free multimodal language model from G​oogle DeepMind that takes text, images and audio and returns text. One renders consequences; the other talks about them. If you arrived looking for a benchmark table that ranks them against each other, none exists and none could — they share no evaluation. What does exist is a real decision, because the two are increasingly deployed in the same pipeline: the reasoner writes the plan, the simulator shows you what the plan does. So the useful question is not which is better but which one your problem actually needs, and what each costs you to hold.

De mismatch, eenvoudig gesteld

G​emma 4 12B has a public evaluation table and PAN 2 does not, and that asymmetry is not a quality signal — it is a category signal. G​oogle publishes MMLU Pro, GPQA Diamond, AIME, LiveCodeBench and a dozen other text and vision benchmarks because language models are measured on a shared yardstick that has existed for years. PAN 2's card publishes Physics-IQ Verified, PhyGround, RBench and a human preference study because video world models are measured on instruments that were invented recently, mostly by the labs building the models.

• Parameters — PAN 2: 29B in a single multi-modal diffusion transformer, with a frozen Qwen3.5-9B text encoder and a Wan 2.2 VAE loaded alongside. G​emma 4 12B: 11.95B, 48 layers, encoder-free, with raw image patches and audio waveforms projected straight into the decoder's embedding space.

• Inputs and outputs — PAN 2: an optional first frame plus a structured world-state caption, out comes an MP4. G​emma 4 12B: text, image, audio and video in, text out, with a 256K-token context and native system-prompt support.

• Licence — PAN 2: Apache-2.0 on its own code and weights, with third-party components under their own terms. G​emma 4 12B: Apache-2.0 under the G​emma 4 licence terms.

• Weights on disk — PAN 2: 86.17 GB across sixty files in the IFM/PAN2 repository, of which 67.37 GB is the transformer itself and 15.87 GB the text encoder. G​emma 4 12B: about 12 billion parameters of BF16, roughly a quarter of the transformer alone.

Waar elk ervan eigenlijk goed in is, afgaand op zijn eigen bewijs.

G​emma 4 12B's numbers are the more legible of the two and they are respectable for the size. Vendor-reported MMLU Pro of 77.2%, GPQA Diamond of 78.8%, AIME 2026 without tools at 77.5%, LiveCodeBench v6 at 72.0% and a Codeforces rating of 1659 put it well clear of the previous generation — G​emma 3 27B scores 67.6%, 42.4%, 20.8%, 29.1% and 110 on the same rows. On vision it posts MMMU Pro 69.1% and MATH-Vision 79.7%. On long context, MRCR v2 with eight needles at 128K lands at 43.4%. Those are G​oogle's own evaluations of G​oogle's own model, and the card says so; treat them as a well-documented vendor claim rather than an independent verdict.

PAN 2 has nothing comparable and does not pretend to. Its claim is narrower and stranger: that a diffusion model conditioned on a structured, time-grounded description will simulate the physical and causal consequences of an action well enough for an agent to compare candidate plans before committing to one. The support for that claim is a 300-case single-action study across six capability categories, a 200-prompt multi-step study with 456 steps, and a 1,400-comparison blind human preference test. All three are the team's own. None of them transfer to anything G​emma 4 12B is measured on.

De vraag over de uitrol is waar de vergelijking pijn doet.

Hier zijn de twee modellen niet langer onvergelijkbaar, omdat beide moeten draaien op hardware die je zelf betaalt.

G​emma 4 12B runs on a single consumer or workstation GPU at BF16 and on considerably less with quantization. It has day-one support in the standard inference stacks, a 262K vocabulary, and a documented set of sampling and thinking-mode defaults. If your workload is document understanding, tool calling or reasoning over a long context, this is a boring, well-trodden deployment and the interesting decisions are all about prompting.

PAN 2 heeft geen saaie deployment. De headline van het vLLM-recept zelf is dat dertien seconden 704×1248-video ongeveer 119 seconden kost op vier B300-GPU's, met classifier-free guidance geparallelliseerd over twee ranks en Ulysses-sequentieparallelisme over twee. De standaard is acht H200's in BF16. De sampler draait vijftig stappen met guidance bij elke stap — honderd forward passes door een 29B-transformer per clip — daarom is de parallellisatie niet optioneel. En het conditieformaat is geen vrije tekst: het model verwacht een gestructureerde world-state-caption, en de repository levert een prompt-rewriter-skill met een validator die een zachte limiet van 1.480 tokens en een harde capaciteit van 1.536 afdwingt tegen het budget van de tekstencoder.

That last detail is the one that decides most evaluations. You cannot type a wish into PAN 2 and get a video. You have to describe events in order, bind actions to actors, and specify camera motion, and the model card is explicit that the agent case studies relied on hand-refined prompts. G​emma 4 12B will happily take a paragraph of prose and give you something useful back on the first try. If you want to know how PAN 2 feels in practice, the honest answer is that it feels like a model with an interface contract, not a chat box.

Waar de twee in hetzelfde systeem thuishoren

The reason this comparison keeps coming up is that agent work now needs both halves. An agent that plans in text and acts in a real environment needs a reasoner to decide what to do and, increasingly, a simulator to preview what doing it would look like. G​emma 4 12B can be the reasoner — it is small enough to co-locate with the rest of a stack, it handles images and audio as inputs, and its instruction-following and tool-use scores are the point of the model. PAN 2 can be the simulator, and its own architecture assumes something like G​emma is already in the loop: the conceptual reasoner is defined as any instruction-following language model, and the report ranks four of them by how often a simulated plan completes end to end — Claude Opus 5.5 first, then GPT-6 Astra at medium effort, then Kimi K3, then Qwen3.8-Flash.

If that is the shape of your system, the pairing is not PAN 2 versus G​emma 4 12B at all. It is PAN 2 plus a reasoner, and the choice of reasoner is where the cost and the quality variance actually live.

Dat is een plek waar een router echt werk doet in plaats van marketingwerk. De simulatiehelft moet op je eigen GPU's draaien — niemand host PAN 2, MBZUAI inbegrepen — maar de redeneerhelft kan een API-aanroep met facturering per gebruik zijn, en wisselen tussen Claude Opus 5.5, GPT-6 Astra en Qwen3.8-Flash om te bepalen welke je plannen het vaakst voltooit, is een routeringsvraag, geen inkoopvraag. Qwen3.8-Flash staat op OrcaRouter, dus je kunt drie planners benchmarken tegen je eigen PAN 2-deployment via één API-sleutel met automatische failover, tegen de lijstprijs van de provider met 0% markup doorgegeven. OrcaRouter host PAN 2 niet en kan het niet routeren. Het kan de goedkope helft van de pipeline elastisch houden terwijl de dure helft aan hardware vastgepind blijft.

Wat zou deze vergelijking veranderen?

Twee dingen, die op 11 november 2026 geen van beide bestaan. Ten eerste: een onafhankelijke evaluatie van PAN 2 door iemand die niet de auteur ervan is — tot die tijd zijn de beweringen over de mogelijkheden door de leverancier gerapporteerd en moeten ze ook zo worden gelezen. Ten tweede: een gehost endpoint of een echte release note voor PAN 2 — het enige bericht op het X-account van het lab is dat niet — waardoor een beslissing over zelf hosten een prijsvergelijking zou worden en de deploymentsectie van deze pagina van de ene op de andere dag achterhaald zou zijn.

G​emma 4 12B needs neither. It is a shipped, documented, widely deployed model with over 1.8 million monthly downloads on Hugging Face, and its evaluation table has had months of community use to find the soft spots. The risk profile of adopting it is close to zero. The risk profile of adopting PAN 2 is that you spend a week on infrastructure for a model whose limitations section says about two in three simulated plans contain at least one step that is not carried out. That sentence is not a reason to skip it. It is a reason to know what you are buying before you rent the GPUs.

Hoe te kiezen

If you need text, images or audio understood and answered, PAN 2 is not a candidate and no amount of GPU budget makes it one. Take G​emma 4 12B, or one of the larger G​emma 4 variants if the 12B's numbers are not enough. If you need to see what an action would look like before you take it — robotic manipulation, egocentric task simulation, physical-reasoning evaluation — G​emma 4 12B cannot do that either, and PAN 2 is one of the few open-weight models that tries. If you need an agent that does both, the answer is a stack, not a model, and the interesting engineering is in the interface between them: the structured caption format that PAN 2 demands is exactly the artifact a reasoner like G​emma 4 12B would have to be taught to produce.