Average Centipawn Loss Explained For Beginners

If you have tried to analyze one of your own games with a chess engine before, you might have come across the term centipawn loss.

The centipawn can be used to evaluate the accuracy of your moves compared to the best move in the position.

A low average centipawn loss indicates very good play, whereas a high value shows you played a lot of inferior moves.

Let’s take a closer look at what a centipawn is and how they are used in chess.

How chess positions are evaluated

Any beginner knows that each piece is worth a certain amount of points. A pawn counts as 1 point, knights and bishop count as 3 points, etc.

So whenever you look at a position and want to determine who is better, you start by comparing the pieces that each player has.

For example, the position below looks quite complicated. However, by quickly counting all the pieces, you can quickly see that white has a rook (5 points) for a bishop and pawn (4 points).

So we can be fairly sure that white has the advantage in this position.

However, determining who is better becomes a lot more difficult if the material is the same and you have to consider more subtle differences such as space advantage, pawn structure, piece activity, development, etc.

Is faster development worth a pawn? Is a strong knight on the 6th rank worth a rook? Are double pawn worth 1 or 2 pawns?

In the position below it’s a lot more tricky to determine who has the advantage. The material is even, and both the white and black position have some pros and cons.

While evaluating these types of chess positions can even be difficult for experienced players, chess engines have no problem determining who is better in a position.

By calculating all possible moves and referencing their databases, chess engines can quickly give the position a score.

Below you can see what the engine evaluation looks like on chess.com. But pretty much any engine or online chess platform office computer analysis nowadays.

As you can see, the engine determines the position is worth -2.01, -3.59, or -4.59 depending on which of the three moves white plays.

Engine scores might seem a little strange the first time you see them.

The scores normally range from -9.00 to 9.00 in which a score 1.00 indicates white has an advantage worth a pawn, while a score of -3.00 shows black has an advantage worth 3 pawns or a minor piece.

The strange part is that computer evaluations are given with 2 decimal numbers. So what does a score of 0.10 or 0.35 even mean?

What is a centipawn in chess?

The centipawn is a unit in chess that is used to determine advantages that are smaller than a pawn. One centipawn equals 1/100 of a pawn and 100 centipawns are worth 1 pawn.

For example, if a chess engine give the position a score of 0.25, that means white has an advantage worth 25 centipawns or 1/4 of a pawn.

In practice, centipawns are only used by chess engines since it’s impossible for regular players to accurately evaluate a chess position up to a single centipawn.

Normally players simply say they have a “small advantage”.

What is centipawn loss?

Centipawn loss is the number of centipawns that a specific move deviated from the most accurate move according to a chess engine.

So if you play the same move as the engine suggested, the centipawn loss will be 0. If you play another move that is also good but not the best, your centipawn loss might be around 20.

If you make a terrible move, your centipawn loss might be 100 or even higher.

Below you can see another screenshot I took from chess.com. On the right you can see all the engine lines.

In this example, your centipawn loss is 0 if you play the engine line Nd5. The second-best move is Bxf6 which gives a centipawn loss of 44-36 = 4.

In general, you want to play moves that have a small centipawn loss, because that shows that the engine agrees that you are making good moves.

However, obsessing about minimizing your centipawn loss isn’t good either!

There is no problem with playing moves that result in a centipawn loss of 10 or 20.

They might not be the best move, but for the average player such tiny deviations are meaningless. And let’s not forget, that chess engines aren’t perfect either.

Just make sure you don’t make moves that lead to huge centipawn loss.

In the position above, playing Ncb1 would result in a centipawn loss of 171, which would be a clear blunder!

What is average centipawn loss?

The average centipawn loss (ACPL) indicates how many centipawns you lost on average during an entire game compared to the best moves calculated by a chess engine.

So while centipawn loss can be used to determine how good a specific move is, average centipawn loss can be used to determine how good you played the entire game.

The lower your average centipawn loss, the better you played.

However, it’s important to note that average centipawn loss is not a perfect reflection of how well someone played.

In long endgames or drawish positions it is much easier to make moves which lead to low centipawn loss compared to complex positions with lots of tactics.

The position shown below is a complete draw. And as the engine lines show, almost any move you make results in a centipawn loss of 0.

However, the game won’t end until both players agree to a draw or the 50 move rule is reached. So you can simply move your king back and forth a few times and lower your average centipawn loss by doing so.

What is a good average centipawn loss?

As I mentioned above, your average centipawn loss can change a lot depending on the type of game you played.

An average centipawn loss of 50 would be good, while anything below 20 would be excellent if you played a relatively quiet and short game

If you played a game with a long drawn out endgame, you can aim for an ACPL below 50. While an ACPL above 75 can still be really good for a sharp tactical game.

Average centipawn loss to Elo rating

A low average centipawn loss indicates that you played an accurate game. And higher rated players normally play more accurate than lower rated players. So you might expect there to be a link between someone’s Elo rating and their average centipawn loss.

But as I mentioned before, there are a lot of factors that influence your average centipawn loss besides your rating. So the correlation between average centipawn loss and rating is very weak.

Patrick Coulombe from chessdigits.com performed a detailed analysis using the lichess.org database.

As you can see form the graph above, it impossible to give an accurate estimation of someone’s rating based on their average centipawn loss and vice versa.

Centipawn loss of grandmasters

The average centipawn loss can vary a lot from grandmaster to grandmaster. And the same grandmaster can have big swings in their average centipawn loss from game to game.

However, you can expect a normal grandmaster to have an average centipawn loss below 40 for most of their games.

Of course, grandmasters that have a more aggressive style will have a higher average centipawn loss compared to grandmasters that prefer to play quiet and strategic positions.

In the 2021 world championship matches, both players had an average centipawn loss below 20 for most of their games.

Unfortunately, Ian Nepomniachtchi blundered in several of his games. Which explains the high average centipawn loss of games 6, 8, and 9.

Centipawn loss and cheating

There is a very weak link between someone’s rating and their average centipawn loss. For example, if you have a 1400 rating your average centipawn loss might vary somewhere between 50 and 100.

That’s a very big range.

But an average centipawn loss below 25 for a 1400 rated player would definitely raise some eyebrows.

Especially if that player manages to play so accurately during multiple games in even very tactical positions.

That’s why many online chess platforms nowadays use average centipawn loss to screen for cheaters.

If someone has an unusually low average centipawn loss for their rating, that might mean someone is using a chess engine to cheat.

Although it’s not a perfect way to prevent cheating, it’s an effective method to block the most obvious cheats.

Centipawn loss and rating inflation

If you have followed the chess scene for a while, you might have noticed that the number of grandmasters and super grandmasters has gone drastically over the last few decades.

Back in 1983, Anotoly Karpov was the online player rated above 2700. Today there are more than 30 players that have passed the 2700 threshold.

This has led many people to believe that there might be Elo inflation. Meaning that a player rated 2700 now would be weaker than a player rated 2700 from 50 years ago.

However, this excellent article by ChessBase actually suggests that the opposite is true!

They analyzed the average centipawn loss of players rated above 2400 over the last 20 year.

The graph above shows that the average centipawn loss of strong players is actually going down. This would suggest that there is rating deflation instead of rating inflation.

It’s nice to know that the current grandmasters are indeed great at the game. And wouldn’t suffer horrible loses if they were to play their predecessors.

Summary

A centipawns equals 1/100 of a pawn, and they are used by chess engines to evaluate positions and the accuracy of moves.

The average centipawn loss of a game is a good indication of how accurately you played.

However, some positions are much easier to play correctly than others. So it’s difficult to say what average centipawn loss would be good or bad without looking at the position.

This also means that the correlation between someone’s Elo rating and their average centipawn loss is very weak.

That being said, if someone has an exceptionally low average centipawn loss that might suggest that they are cheating.

The average centipawn loss of grandmasters has also gone down over time. Which suggest that there is rating deflation instead of rating inflation.