Fewer guesses, expensive guesses: what it really costs to search for a Minecraft seed on a quantum computer
Researchers designed a quantum search for a Minecraft world’s seed, counted what it would cost to run, and tested a tiny version on IBM and Quantinuum machines. The math works. The cost doesn’t yet. The full search would need at least 1.65 billion two qubit gates before error correction, while a 2012 desktop chip checks every candidate in about 16 seconds.
A treasure hunt with a known answer
Every Minecraft world is generated from a single number called the seed. The same seed always produces the same world, with the same shipwrecks in the same places. If you know where a few shipwrecks are, you can work backward and test candidate seeds until one puts them exactly there. Free software already does this. The authors took its final step, a search over about 4.3 billion candidates, and asked what it would look like on a quantum computer.
They did it in three steps. First they designed the full quantum circuit that runs the game’s random number generator on a candidate seed and checks the shipwreck positions. Then they counted its gates with a compiler, which gives an estimate; the full circuit was never run. Finally they shrank the same check from 48 bit numbers to 3 bit numbers, leaving 8 candidates and one right answer, and ran that tiny version on real hardware.
Two words worth knowing
A two qubit gate is one basic step in which two qubits interact. It is the slowest, most error prone step, so it is the standard measure of cost.
An oracle is the part of a quantum search that checks whether a guess is right. Here, it is the game’s random number generator plus the shipwreck check, built from quantum gates.
Why nobody ran the real thing
Current machines are many orders of magnitude short of running a circuit with over a billion two qubit gates. That takes error corrected machines, which don’t exist at this scale yet.
8,461 rounds instead of 4.3 billion
A normal computer tries candidates one by one. Grover’s algorithm, the famous quantum search, needs roughly the square root of that many rounds, so here it would take 8,461 rounds instead of 4.3 billion checks.
To check a guess, the quantum computer has to run the game’s random number generator inside the circuit.
The authors’ first version of that check needed about 2.04 million two qubit gates per round. They optimized it down to 48,896, a cut of about 42 times. Over 8,461 rounds, that is about 1.65 billion in total, before the rest of each round, fitting the circuit onto a real chip, and error correction.
The same budget could have bought something useful. In 2018, Andrew Childs and colleagues at the University of Maryland estimated that simulating a chain of 50 interacting magnetic particles, a standard model of magnetic materials at a size ordinary computers struggle with, needs about 180 million two qubit gates by their most cautious count. The Minecraft budget covers that roughly nine times. Breaking 2048 bit RSA encryption is another story. Craig Gidney’s 2025 estimate needs about 6.5 billion Toffoli gates, a heavier three qubit step that takes several two qubit gates to build, which makes it roughly 20 times the Minecraft search. Both comparisons are order of magnitude only, since the papers count gates in different ways.
What the Minecraft budget could buy. RSA is counted in Toffoli gates, a heavier unit. Sources: Reeves et al. 2026; Childs et al. 2018; Gidney 2025.
What the tiny version showed
The tiny test on Quantinuum H2 against a perfect machine. Source: Reeves, Sopkin, Searles, arXiv:2610.33139 (CC BY 4.0), Table V.
Each Grover round nudges the odds toward the right answer. The odds rise up to an ideal number of rounds, then fall again. In the tiny version, the ideal was two rounds. Quantinuum’s H2 found the right answer about 73% of the time after one round, near the ideal 78%. At two rounds it stayed at 73%, short of the ideal 95%. On IBM’s Heron chips, all six test sessions hit their best odds after one round, a round early, at only about 15 to 18%, barely above the 12.5% of a blind guess. The paper only reports the early peak, and I read it as noise.
What this result does not say
The 1.65 billion is a compiler count, not a measurement, and covers only the check.
This is a university preprint, not peer reviewed, and not company work. IBM and Quantinuum only provided hardware access.
An old warning, now with a number
Researchers at Google and Microsoft (Babbush et al., 2021; Hoefler, Häner and Troyer, 2023) have argued that square root speedups like Grover’s are unlikely to pay off in practice, because each quantum step is so much slower than a classical one. This preprint puts a real gate count on that argument.
What I take from it
Fewer guesses don’t help much if each guess is expensive. The question is not how many rounds, but what one round costs.
I suspect Grover style gains arrive late on vendor roadmaps, after chemistry and simulation, because the cost per round eats most of the edge until machines are both error corrected and fast.
Sources
Reeves, Sopkin, Searles, “The Cost of a Real Grover Oracle: Minecraft Seed Search from Software to Quantum Hardware”, arXiv:2610.33139. https://arxiv.org/abs/2610.33139
Hoefler, Häner, Troyer, “Disentangling Hype from Practicality: On Realistically Achieving Quantum Advantage”, Communications of the ACM 66(5), 2023. https://doi.org/10.1145/3571725
Childs, Maslov, Nam, Ross, Su, “Toward the first quantum simulation with quantum speedup”, PNAS 115(38), 9456 (2018). https://arxiv.org/abs/1711.10980
