The short answer
We solved every legal move to the end of the game in an endgame with two corners open. The best move was the corner a1 (+14); the other corner, h8, finishes ±0 — 14 discs apart. Two corners are not worth the same. Across 400 positions, taking a corner was best 50.8% of the time.
What we are testing
"A corner is a stable disc that can never be flipped, so take it if you can" is Othello's iron law. But does that law still hold when two corners are open at once? In the endgame we can solve the position completely, so let's find out.
How we tested it
Every number here is output from our own solver (negamax with alpha-beta). We take an endgame position with few empty squares, play each legal move in turn, and solve what follows to the very end of the game. There is no evaluation function and no approximation, so the margins are exact rather than estimated. One position cannot show how typical something is, so each article solves three.
Position by position
Black has six legal moves, and two of them (a1 and h8) are corners. Solving every move to the end of the game gives this (margin is Black's final disc differential, best play by both sides):
| Move | Discs flipped | Final margin | Type |
|---|---|---|---|
| a1 | 3 | +14 | corner |
| c7 | 3 | +6 | — |
| a7 | 2 | +4 | C-square |
| h8 | 6 | ±0 | corner |
| a6 | 4 | ±0 | — |
| a5 | 5 | ±0 | — |
Where the difference comes from
The two "corners" turn out to be very different. a1 finishes +14, a clear win, while the other corner, h8, finishes ±0 — a draw. Take the wrong corner and you drop a win to a tie.
There is more: h8 is also the move that flips the most discs (six) in this position. The six-disc corner is far worse than the quiet three-disc a1; once again, disc count is no guide to the result. Taking a corner matters less than what happens all the way to the end after you take it.
Exceptions
In this position Black cannot actually lose: even the worst move finishes at ±0, turning a win into a draw. Sometimes what is at stake is the size of the win, not the result.
That does not make the corner a safe default. Taking the big-looking h8 drops Black to ±0 - below c7 (+6) and a7 (+4), neither of which is a corner. With two corners open, which one you take has to be read out.
Playing the best move
Watch the best move, a1, being played: the disc goes down first, then the captured discs turn over one at a time. The gold frame marks the square played; the gold rings mark the discs that flipped.
As stated in our editorial policy, the numbers are the real computed results, printed as-is.
The same question over 400 positions
The positions above are solved one move at a time. The same question was also counted over 400 random positions (10 empty squares, 2617 moves solved in total).
A corner was legal in 246 of them, and taking it was best in 125 (50.8%) — about half. Always taking the corner when one was available would cost 5.11 discs on average: small next to the other rules of thumb, but not zero.
What this does and does not show
- It shows: for this position, an exact solve of every legal move to the end of the game. Not an approximation.
- It does not show: anything about every position — 1 is a very small sample. For a wider count see the 400-position study.
- The assumption: the margins assume best play from both sides to the end.
- Reproducing it: the positions and results are in
assets/othello-analysis.json, regenerated bytools/generate-othello-analysis.mjs.
What to remember at the board
- Two corners, two values: a1 finishes +14, h8 ±0
- Over 400 positions, taking a corner was best 50.8% of the time (of 246 where one was available)
- With two corners open, read out which one to take
- The question is not whether to take a corner but which corner
Try it yourself
You can set these positions up and play them out. The evaluation bar moves after every move, so the difference between a good move and a poor one shows up as you play.
Check the same position yourself
Set up the endgame and see how the choice of corner changes the result. Free, no account.
⚫ Enter the Dojo