Last updated:
Average chess accuracy rises with rating, but slowly. In GM Larry Kaufman's study of chess.com 10+0 rapid games, players rated 1000 averaged 75.66% accuracy, players rated 1600 averaged 80.55% and players rated 2000 averaged 83.93%. As a rule of thumb, one point of accuracy is worth roughly 100 rating points.
| Rating (chess.com rapid) | Average accuracy |
|---|---|
| 400 | 68.83% |
| 1000 | 75.66% |
| 1600 | 80.55% |
| 2000 | 83.93% |
| 2400 | 89.19% |
The rest of this article explains where these numbers come from, how to read your own accuracy against them, and why a single game says much less about your level than it seems to.
How the data was measured
The table comes from Accuracy And Ratings On Chess.com, a study GM Larry Kaufman published on the chess.com blog in August 2023. He used 10+0 rapid games played on chess.com and the accuracy that chess.com's own Game Review computes for them. That accuracy is produced by CAPS2, chess.com's proprietary scoring system.
We are not publishing a dataset of our own. ChessRank does not collect anonymized statistics about the games people analyze, so we have no table of accuracy by rating built from our users, and we will not invent one. If that changes, this article will say so and show the sample size behind every number.
Two limits follow from the source. First, the study covers 10+0 rapid only. Second, the numbers are chess.com accuracies. Other sites, ChessRank included, compute accuracy with their own engine settings and formulas, so their scale is close but not identical.
Accuracy by rating band
Kaufman also gave a simple approximation that fits his rapid data: Rapid Accuracy = Rapid Elo / 100 + 64. It is easy to apply to any rating band. The table below lists the formula next to the measured averages where the study has them.
| Rating band | Formula estimate | Measured average |
|---|---|---|
| 400 | 68% | 68.83% |
| 600 | 70% | |
| 800 | 72% | |
| 1000 | 74% | 75.66% |
| 1200 | 76% | |
| 1400 | 78% | |
| 1600 | 80% | 80.55% |
| 1800 | 82% | |
| 2000 | 84% | 83.93% |
| 2200 | 86% | |
| 2400 | 88% | 89.19% |
The formula is an approximation. Kaufman notes that when accuracy is below 80, it overstates the rating by about 100 Elo. In other words, in the lower bands a given accuracy belongs to a slightly weaker player than the formula suggests.
Blitz, bullet and classical
The study does not include blitz, bullet or classical games, and we have not found a published table of the same quality for them, so we do not quote numbers. Less time per move usually means more mistakes, so it is reasonable to expect lower accuracy at the same rating in blitz and bullet. How much lower is not something this data can tell you. Compare yourself only with numbers from the same time control and the same site.
How many Elo points is one accuracy point?
The formula says one accuracy point is worth 100 rating points. The measured averages show that the exchange rate is not constant. Dividing the rating gap by the accuracy gap between neighbouring rows of Kaufman's table gives:
| Rating range | Accuracy gained | Elo per accuracy point |
|---|---|---|
| 400 to 1000 | 6.83 | about 88 |
| 1000 to 1600 | 4.89 | about 123 |
| 1600 to 2000 | 3.38 | about 118 |
| 2000 to 2400 | 5.26 | about 76 |
So between 1000 and 2000, improving your average accuracy by one point corresponds to more than 100 rating points. These are averages over many games. They describe a population, not a promise for your next game.
White vs Black
In Kaufman's data White averages about 1.12 accuracy points more than Black. White moves first, gets the initiative more often and makes fewer forced concessions. When you compare your games, compare your White games with each other and your Black games with each other, or allow for that gap.
Why your accuracy swings from game to game
Accuracy measures how much winning chance each move gave away. That makes it sensitive to the kind of game you played, not only to how well you played it.
- Decided positions. Winning chances barely move once a position is clearly won. With the published Lichess formula, going from +10 to +7 costs about 4.6 points of winning chance and still scores about 81% for that move. Giving away the same three pawns from an equal position costs about 25 points and scores about 31%.
- Game length and opening moves. A short game with many book moves has few moves where you can go wrong. A long, complex game has many.
- Forcing lines. Recaptures and checks with one sensible reply are easy to play accurately and pull the average up.
- Opponent strength. A weaker opponent is more likely to hand you a clearly winning position, and a clearly winning position is easier to play at high accuracy.
- Time pressure. Moves played with seconds on the clock are where most large drops happen.
This is why the same player can score 92% in one game and 71% in the next. Kaufman's study also shows how much data a reliable estimate needs: with about 50 competitive games per player, the median error when predicting the ratings of top players from accuracy was 48.5 Elo. One game is nowhere near that.
How this compares with ChessRank
ChessRank's Game Analysis runs Stockfish in your browser and turns the engine's evaluations into accuracy with its own model, calibrated against chess.com reviews of real games. The scale is meant to be close to chess.com's, but the two are different systems with different engines and settings, so the same game can differ by a few points. Use the table above as a guide to chess.com numbers, and compare your ChessRank accuracy with other ChessRank games.
The most useful number is your own trend. The Report analyzes your recent games in one batch and shows your average accuracy per time control, and Chess Stats tracks how it moves. The method behind every number is explained in How Is Chess Accuracy Calculated?, and the per-game strength estimate in Chess.com Game Rating Explained. For a short answer on what to aim for, see What Is a Good Accuracy Score in Chess?
Frequently asked questions
Is 80% accuracy good in chess?
In GM Larry Kaufman's study of chess.com 10+0 rapid games, 80.55% was the average accuracy of players rated 1600. So 80% in a rapid game is about what an average 1600 player produces on chess.com. In blitz or bullet the same score is likely worth more, because faster games bring more mistakes.
What accuracy do grandmasters have?
Kaufman's table ends at 2400, where the average rapid accuracy was 89.19%. The study does not give a separate figure for grandmasters. Even very strong players have single games far below their average, because accuracy depends on how sharp the game was.
Can I convert accuracy to Elo?
Only roughly. Kaufman's approximation is Rapid Accuracy = Rapid Elo / 100 + 64, so 78% maps to about 1400. He notes that below 80% accuracy this overstates the rating by about 100 Elo, and it only fits chess.com rapid accuracy averaged over many games.
Why is my accuracy higher against weaker players?
A weaker opponent is more likely to give you a clearly winning position. Once a position is clearly won, most moves cost very little winning chance, so they score high even when they are not the best move. Accuracy measures lost winning chances, not difficulty.
Why is my chess.com accuracy different from ChessRank?
Chess.com computes accuracy with CAPS2, which is proprietary, on its own servers. ChessRank runs Stockfish in your browser and uses its own model calibrated against chess.com reviews. The scales are close but not the same, so a difference of a few points on one game is normal.
Is accuracy lower in blitz than in rapid?
Usually, yes, because there is less time for each move. Kaufman's study only covers 10+0 rapid, so there is no published number for how much lower. Compare your blitz games with your other blitz games.
Sources
This article uses GM Larry Kaufman, Accuracy And Ratings On Chess.com, chess.com blog, August 2023, for every rating and accuracy figure, the approximation formula and the White and Black gap. The illustration of decided positions uses the published formulas on lichess.org (lichess.org/page/accuracy). Facts about ChessRank come from its own code. ChessRank is an independent project and is not affiliated with, endorsed by or sponsored by Chess.com or Lichess.