What is a good accuracy in chess?
There's no official "good" accuracy in chess. Chess.com says most of its accuracy scores fall roughly between 50 and 95, and Lichess warns that comparing accuracy to a school grade is flawed. A more useful benchmark is your own: in 1,470 of my games (almost all rapid), my median Chess.com accuracy was 80.5% in wins and 68.0% in losses.
So "good" depends on the site, on how hard the position was and on what your opponent did. Below are the official definitions, my real numbers by result and rating, and why a 93% game can still be a loss. (I'm nacsol3, an adult improver who went from 631 to 1385 in rapid on Chess.com.)
What a good accuracy looks like in real games
Numbers from my games of at least 20 moves each side, January 2024 to October 2026, almost all rapid. The Chess Coach column uses the Lichess formula; the Chess.com column is the score Chess.com reports, available for 610 of these games.
| My games | Chess Coach accuracy (median, middle half) | Chess.com accuracy (median, middle half) |
|---|---|---|
| Wins | 87.1% (78.9–93.0) | 80.5% (74.8–85.9) |
| Draws | 78.5% (72.3–83.8) | 72.7% (69.0–76.9) |
| Losses | 73.4% (61.9–80.1) | 68.0% (62.6–74.2) |
| Rated below 1000 | 78.6% | 77.3% |
| Rated 1000–1199 | 80.5% | 77.0% |
| Rated 1200 and above | 79.9% | 76.1% |
"Middle half" means half of the games fall inside that range. Three things stand out:
- Wins and losses overlap. Plenty of my losses scored higher than my average win.
- Accuracy didn't grow with my rating. It stayed near 77–80% from below 1000 to above 1200, because my opponents got stronger too.
- Blunders matter more. In games with no blunder I won 93.4% of the time (197 games); with one blunder, 63.0% (300 games); with two or more, 37.0% (973 games). A blunder here is a move that drops 15 or more points of winning chance (the three error labels explained).
These are one player's numbers, not a rule for everyone.
How Chess.com and Lichess define accuracy
Chess.com. Its Help Center article How is accuracy in Analysis determined? (dated June 21, 2025, checked November 2026) says accuracy measures how closely you played to the engine's best play, and the closer to 100, the closer to "perfect" play. The score now uses CAPS2, a revised version of its Computer Accuracy Precision Score, designed to feel like a test grade, with most scores between 50 and 95. Chess.com doesn't publish the full formula.
Lichess. Its accuracy page (checked November 2026) publishes the method. Each evaluation becomes a Win%, your chance of winning from that position, and each move gets an accuracy from how much Win% it lost. The game score combines a volatility-weighted mean with a harmonic mean, so a few bad moves pull it down hard. The same page says "there is no universally accepted way to compute accuracy percentages," and that the complexity of a position affects accuracy.
That's why the two numbers rarely match. In the 610 games where I have both, the Lichess-style score was a median 5.5 points higher. The Lichess vs Chess.com review comparison has a full game scored both ways.
Why 93% accuracy can still lose
March 13, 2026. I'm White (1228) against ihabbbz (1250) in a 10-minute London System (game on Chess.com). My accuracy for the game was 93.3%, excellent by any standard. I lost by checkmate. My opponent's accuracy was 98.8%.
The engine wanted the pawn to b4, attacking Black's queen, which had no defender, and keeping things only slightly worse. I pushed the other pawn to c4 instead. Black's best reply is to push the centre pawn to d4, and the position became clearly better for Black. My win chance fell from 43.9% to 26.8%. That move is marked as a blunder. On the next move I made a mistake, and my opponent didn't let go.
Three marked moves out of 38 (an inaccuracy, this blunder and a mistake) barely dent an average. They were enough to lose. Accuracy tells you how clean the whole game was, not which move decided it. I wrote more about this in my week with my best accuracy and a 5–9 score.
How to use your accuracy score
- Compare with yourself, on one site. Track your own Chess.com or Lichess number over many games, never one against the other.
- Look at the result and the blunders first. One blunder outweighs ten small inaccuracies.
- Check the opponent's number. 85% against an opponent's 95% is a different story from 85% against 60%.
- Ignore very short games. A resignation after five moves can show 99%.
If you want the per-move number behind the score, see what centipawn loss means.
How Chess Coach shows accuracy
Chess Coach is a connector for Claude and ChatGPT that reviews finished Chess.com and Lichess games with Stockfish on our own server. It computes accuracy with the Lichess Win% formula, so expect it to differ from Chess.com's number. The review focuses on the key moments instead: the moves that changed the evaluation most, on an interactive board with a red arrow for your move and a green one for the better move, explained in plain words. The free plan includes one in-depth review a day. Connect Chess Coach and ask it to review your last game.
Frequently asked questions
Is 80% accuracy good in chess?
Neither Chess.com nor Lichess sets a cutoff. In my own 1,470 games, almost all rapid (rated 707 to 1385), my median Chess.com accuracy in wins was 80.5%, so for me 80% on Chess.com was a typical winning game.
What accuracy do beginners usually get?
Neither site publishes an average by rating. Chess.com only says that most scores under its current CAPS2 system fall roughly between 50 and 95.
Why did I lose with 90% accuracy?
Accuracy averages all your moves, and one blunder can decide the game. Your opponent may also have played even better: in one of my losses I had 93.3% and my opponent had 98.8%.
Is Lichess accuracy the same as Chess.com accuracy?
No. Lichess publishes its formula and says there's no universally accepted method; Chess.com uses its own CAPS2 algorithm. In 610 of my games with both numbers, the Lichess-style score was a median 5.5 points higher.
Does higher accuracy mean a higher rating?
Not directly. My median accuracy stayed around 77–80% in every rating band from below 1000 to above 1200, while my rating doubled. What changed was how often I blundered.