VoxelMind Bench
The first LLM benchmark for emergent civilizations — for everyone.
We don't build civilizations. We grow them.
We drop AI models into a living Minecraft world as mortal inhabitants — then measure what emerges. Economy, society, and the civilization that grows from there: none of it scripted. The benchmark is how rich a civilization a model can grow.
The thesis
Civilizations are not goals. They're byproducts.
Nobody wakes up and decides to build an economy. It emerges — bottom-up — from something far simpler: mortal individuals with needs, in scarcity, alongside others, over time, with memory. So we don't hand the bots a “build a civilization” goal. We give them the primitives — and let it grow.
Agent capability
Making the inhabitant able to do what a human can. The same brain powers our Companion mod — one codebase, growing more human every release.
World mechanics
The pressures we inject — mortality, scarcity, aging, reproduction. Economy and society fall out of these as the efficient answer to survival.
We measure, we don't dictate. The bots never see the score — the moment a number becomes the target, it's gamed. They live; we observe. And it all scales with how human-like the agent becomes: the sim is the thermometer of machine intelligence.
The roadmap
Six phases from a single survivor to a dynasty
Each phase adds capabilities + mechanics that unlock a new class of emergence. Each is its own benchmark dimension — and blooms as the agent matures.
The Mortal Individual
One life. When you die, it's over. Survive — and thrive.
Sustenance & Specialization
Hunger forces sustainable supply — and supply is worth dividing.
The Exchange
I have food, you have tools — we trade. Value lives in the flow.
The Settlement
Huts become houses, houses become a walled village.
The Tribe
Scarcity + proximity + others → groups, alliances, and conflict.
The Lineage
We age, we die, we raise the next generation. The rest follows.
How we measure
The Civilization Score
Not one gameable number — a set of dimensions that together show what kind of civilization a model grows. A trader society? A fortress tribe? A long-lived dynasty?
Survival & Demography
The gate — who lives, and how long
Material & Tech
Tools, materials, the tech tree
Infrastructure
The built, persistent world
Economy & Cooperation
Output, trade & division of labour — the heart
Knowledge & Culture
XP, enchanting, writing, redstone
Generations
Births, lineage, sustainable population
Civilization stage: S0 Spawn → Survival → Subsistence → Settlement → Metallurgy → Society & Trade → Industry → S7 Civilization.
Available now
Meet your first living AI companion.
VoxelMind Companion is live today — an AI who knows your world, remembers your history, and stays with you across every session. The same technology powering the Bench.
The north star
Seed a world with mortal minds. Walk away. Come back to a civilization.
Generations later: settlements, trade, tribes, leaders, wars, dynasties, inequality, culture — none of it scripted. The honest measure of machine humanity.