500-game benchmark

The proof that this Block Blast! solver works: 500 simulated games

Three players, the same 500 games and the same pieces: random lasts 3 turns, the one that only clears lines 20, and this solver 40. With its date and method in plain sight.

Any tips website can say "this is the best move". We publish the number that measures it, how it was measured, and the day it was measured.

The auto-play benchmark is the experiment we used to decide whether this app deserved to be published. It simulates full games of an 8×8 block puzzle and pits three players against EXACTLY the same seeds, meaning they get the same pieces in the same order: a random player that only places wherever it fits, a greedy player that always picks the move that clears the most lines that turn, and this app's solver, which evaluates the board left afterward. Each one plays until none of its three pieces fit, and the turns survived and estimated points are recorded.

The result of the latest run, from November 2, 2026 over 500 games, is this: the random player survives a median of 3 turns, the greedy one 20, and the solver 40. That means the solver lasts twice as long as a player who already plays well but only looks at the present, and more than ten times as long as random. In estimated points, the medians are 430, 5,050 and 10,900. The solver is measured with the SAME budget it has on your phone, 250 milliseconds per turn: if the benchmark gave it more time, it would be measuring a solver you don't have.

The benchmark also produces the part that really makes the solver play well: the weights of its evaluation function. Four different weight sets were tested over the same games, and the one that punishes single-cell holes hardest won. Those are the weights frozen into the app, not a hunch: the move you get when you paint your board is chosen with the numbers that won here.

And it's published inside the app itself, not just on this page: the solver tab carries the "How we know this works" card with the three medians, the number of games, and the run date, plus the assumption we make and can't verify: that the game deals pieces uniformly among the catalog's 33, because Block Blast!'s real distribution isn't public. That's exactly what you won't find on a tips website.

No sign-up

Try 500-game benchmark right here

Loading the tool…

Use it free without an account. If you want to save your result, you can sign up afterwards.

Save / Sign up

Benefits of 500-game benchmark

Same seeds for all three

the random player, the greedy one and the solver get the same pieces in the same order in each of the 500 games. Without that, the comparison would be worthless, because a player could simply have had better luck.

The number is in plain sight, with a date

3, 20 and 40 median turns, run on November 2, 2026, 500 games. It's not "improve your game": it's a figure you can cite, criticize or compare.

Measured with your phone's real budget

the benchmark gives the solver the same 250 milliseconds per turn it has when solving your board. There's no smarter lab version than the one you use.

The opponent isn't a dummy

the greedy player, the one that always clears the most lines on the turn, is exactly the strategy almost every tip online recommends. It lasts 20 turns. Looking at the board that's left afterward is worth 20 more.

The evaluation weights are calibrated, not invented

four weight candidates competed over the same games and the winner was frozen into the app. The one that penalizes single-cell holes the most won, and that's the decision you notice in your moves.

It was a condition for publishing, not a decoration

the benchmark carried a quality gate, and if the solver didn't beat the greedy player by more than 15% in median and in mean, the app wouldn't ship. It beat it by double.

The uncomfortable assumption, stated plainly

we assume uniform piece draw over the catalog's 33 shapes because the game's real distribution isn't public. We say so in the app and here, instead of hiding it behind the number.

And what the benchmark is not

a precalculated figure we publish with its date, not an experiment launched from the app, nor a promise that you'll last 40 turns. It measures the solver's decisions, not your game.

40-turn median. Now try it on your own board

Paint your 8×8 board, pick your three pieces and get the move calculated by the same engine measured in these 500 games. Free, no registration, no usage limit.

Does the Block Blast! solver work? 500 games measured