a dark pond, illustrated in halftone — every ripple surfaces a world. Orange rings the ones that are playing.
Any game, any recorded reality, delivered as a living, playable world model: your data, your weights, in weeks for five figures. Five days after the first open multiplayer world model was released, ours was the only known reproduction serving players.
MIRA, the first open multiplayer world model, was released July 6 with code and data, weights withheld. By July 11 we had trained the codec and the 1B model from the recipe, with the four-player fine-tune training on-curve, and were serving it in a browser (gated) on production session infrastructure. Nobody else has a known reproduction.
What that took: the codec, the 1-billion-parameter world model, and the four-player fine-tune, trained on rented spot GPUs for roughly a hundredth of the original lab's capital. Doing that without a datacenter is fleet-resume protocols, eval discipline, and cost engineering.
What it proves: the recipe is public; the execution is the product. The reproduction is chapter one, never the business. The business is doing this for your game, on your data, and handing you the weights.
For most jobs the real game engine is cheaper and exact — we will tell you when it is. A trained world model earns its keep on the things an engine cannot do:
Trained on real matches, it absorbed how people actually play. The opponents in our demo were never scripted — they are the training data, alive. An engine cannot answer "what would a player do here?"; the model is that answer.
An engine replays a log deterministically. The model takes any real moment and plays it forward differently: alternate futures with realistic opponents in each. Counterfactual replay has no engine analog.
N rollouts from one state give a probability map over futures: win odds by choice, risk by position. Engines compute outcomes; the model estimates distributions.
You cannot backprop through a game engine. Gradients of futures with respect to actions mean agents train inside the model, GPU-batched. That is the property the embodied-AI market pays for.
Engine states must be reachable. The model needs pixels: start play from a screenshot, an edited frame, a composited scene. And it survives its creator: video plus inputs keeps a world playable after the servers are gone.
“The planet in the glass was never the terrain — it’s the population.”
One arm, one car, one player. Conditioning on every participant's actions and attributing consequences correctly is the hard trick. A handful of teams have run it; ours is the only pipeline that delivers it as weights you keep.
The reproduction cost ~$10–15k on rented spot GPUs. The pipeline is amortized; your world rides it. Proven, not promised.
API labs meter imagination, which is ruinous at RL-training scale and impossible where data can't leave the building. We deliver the trained weights on your data, on your cluster if you want. Your simulator is your moat.
Each one survived the filter "why not use the real game?" — the buyers either don't have the engine, or need what it can't do.
The agent-training market hand-builds environments at engineer salaries. We turn recordings of play into GPU-batchable, uniform-API environments grounded in how humans actually behave.
first check $25–100k · fast cyclesGame-agent and QA companies, and studio central tech that already trains replay bots by hand. We are the layer underneath their agents.
pilots $25–150kMarketing moments now ("play the anniversary, remixed"), and a second life for the replay archive: license your world to the AI-training market, rev-share on data you already own.
$50–200k per momentFleets and human-robot floors are multi-agent, which is our native shape. Base models partially serve this lane; the gap is the ops, and delivered weights are the product where data can't leave the building.
expansion act · quarters-long cyclesServing cost is the wall between a world model and consumer scale. Measured at the top, projected below:
Honest evals, the war-story writeup, a playable demo; weights follow the rights conversation. The credential.shipping now
A beloved multiplayer game, trained with its rights holder as the first commercially licensable open multiplayer world model. Not a reproduction: an artifact nobody else has.next
Agents trained inside the world model — the application the MIRA paper names first and nobody ships.brief written
We are not a foundation-model lab and don't pretend to be: the recipe is open and the giants own the capital war. Rights run through the rights holders: reproduction data is non-commercial and stays that way until the people who own it say otherwise. And nobody has bought a per-game world model before, because none has been for sale — the first reference customer is the point of the chapters, and we'd rather tell you that than discover it together later. Consumer retention on dreamed worlds is unproven; we instrument it instead of assuming it.
A pilot starts with recordings — replays, video with inputs, telemetry. You get back a playable world model of your game, evals against the real thing, and the weights. Yours.
the world model foundry · by the team serving world models in production at play.alakazam.gg