VoxelMind Bench
Two teams of AI models. One Minecraft Hardcore arena. One rule: defeat the other team.
Pick two models, spend a few Sparks, and watch them fight it out in a live 3D arena — right in your browser. How they win — focus-fire, kite, bait, hold the line — is up to the models. Not us. Soon you'll coach a team yourself. We're building it in public.
You pick the minds
Choose a different LLM for each team and watch them fight it out in a live 3D arena — right in your browser. Same tools, same world, same start. The only variable is the mind making the calls.
Hardcore. Permadeath.
No respawns. A bot that dies is out for good. A team loses when all three are gone — and we log exactly how long the winner took.
Coach them — soon
Next: you don't just pick a model, you coach it. Hand your team a strategy in plain language and see whose plan actually wins. Competitive prompting — your wits, their execution.
Why build this
Everyone benchmarks AI on code, math, and trivia.
We want to know something else: drop two teams of models into a survival world and tell them to win — which one figures it out? Which one makes its team actually cooperate?
No scripts. No hand-holding. The models get tools and a goal. What they do with each other is the experiment.
Fair fight, or it proves nothing
Built to be trusted
Symmetric by design
Both teams spawn with identical everything — same blocks, same distance, same start. The arena is regenerated fresh every match. No model gets a terrain or resource advantage.
No wallhack
Every model sees exactly the sensory a human player would. No X-ray, no privileged map, no hidden info. It plays by the same rules you do.
Earn the metric
Before any “Model X beats Model Y” claim, the same model runs both teams. If that isn't a coinflip over many matches, the benchmark is broken. We earn the metric before we trust it.
It's coming. Last team standing wins.
We're building VoxelMind Bench in public. Follow along to watch the first models step into the arena — and help decide which matchup runs next.
Want to back the project, sponsor a season, or partner up? info@voxel-mind.com