At what Elo do you stop blundering?
You don’t, at least not in the games we measured. In analysed Lichess blitz games from September 2026, players rated about 2500 on Chess.com’s scale still made 3.0 blunders a game, against 4.1 at 1000, and a 1700 makes only about one blunder fewer every five games than a 1200. Yet by the rating formula a 1700 should score about 95% against a 1200, and a 2500 more than 99% against a 1000. So what separates chess players, if not blunders? We measured it across millions of games from September 2026, from about 500 to 2500.
The numbers in brief
- Blunders a game
In analysed blitz games, players rated about 800 on Chess.com make 4.2 blunders per game; at 1700, 3.8; at 2500 they still make 3.0.
- Lost in the first 10 moves
At 800, about 1 lost game in 7 is decided in the first 10 moves; at 1700, 1 in 15; at 2500, 1 in 41.
- Big blunders a game
Big blunders, worth about a piece, fall from 1.19 a game at 800 to 0.90 at 1700 and 0.60 at 2500.
- Lead changes a game
Between two players at 800 the lead changes hands 1.22 times a game; at 1700, 0.93 times; at 2500, 0.60 times.
- Expected score against a 1200
By the Elo formula, against a 1200 an 800 should score about 9%, a 1700 about 95% and a 2500 more than 99%.
Source: the Lichess open database, September 2026, analysed blitz games, ratings converted to Chess.com’s scale. How we measured, below, has the details.
How much better is a 1700 than a 1200?
Every step up the ladder is a big one. By the Elo formula that rating systems are built on, a 200-point gap already gives the stronger player about 76%, and 500 points about 95%. A small edge, repeated in every game, becomes a lopsided record.
How many blunders per game is normal?
A blunder here is a move that throws away 10 points or more of the player’s winning chances. Players at 600 make 4.3 of them a game, at 1000 4.1, at 1200 4.0 and at 1700 3.8. Even at 2500 it’s 3.0. Research on 200 million online games found the same: “a player rated 1800 will almost always defeat a player rated 1200, yet knowledge of rating is not providing much predictive power in determining blunders on any individual move” (Anderson, Kleinberg and Mullainathan, 2016).
What separates stronger players, if not blunders?
The measures come in different units, so each bar shows how much lower a measure is at 2500 than at 1000, relative to its own value. The blunder count falls by a quarter, from 4.1 to 3.0 a game. Three measures fall much further: the share of lost games decided in the first 10 moves (10.5% to 2.5%), how often the lead changes hands between two players of the same level (1.12 to 0.60 times a game) and big blunders, the ones that throw away about a piece (1.09 to 0.60 a game). The chart shows where the levels differ most, not which skill matters most.
1. Fewer losses decided in the opening
A loss counts as decided in the opening when the winner’s chances reached 70% within the first 10 moves and stayed there to the end. At 600 and 800, about 1 lost game in 7 goes like that, and at 1000 and 1200 about 1 in 10. At 1700 it’s 1 in 15, at 2000 1 in 21 and at 2500 1 in 41. Stronger players’ losses are decided later instead: at 2500, 60% are decided in the middlegame and 26% in the endgame, against 65% and 15% at 1000.
2. Their big blunders are rarer
A big blunder throws away 30 points or more of winning chances, roughly a piece. At 600 players make 1.26 of them a game on average. At 1000 it’s 1.09, and players made at least one in 60% of their games. At 1200 it’s 1.02 a game, at 1700 0.90 and at 2500 0.60, where players made at least one in 39% of their games. The smaller slips stay: blunders from a level position, where both sides still have chances, are about 2 a game from 1000 to 1700. If your big blunders are hung pieces, how to stop hanging pieces shows the three habits behind them, each with the question that catches it before you move.
3. Their games swing less
A side leads once its winning chances reach 65%, and the lead changes hands when the other side gets there. Between two players at 600, that happens 1.29 times a game. At 1000 it’s 1.12, and 57% of games see at least one lead change. At 1700 it’s 0.93 times a game, at 2000 0.81 and at 2500 0.60, where 38% of games change hands at least once.
4. They cash in on blunders
A chance to punish is the move right after an opponent’s blunder, while the game was still open. It counts as punished when the player keeps at least half of what the blunder gave them, and as blundered back when the reply is a blunder itself. At 600 players punish 60% of these blunders and blunder straight back 37% of the time. At 1000 it’s 62% and 35%, still about one time in three. At 2500 it’s 70% and 26%, about one time in four. These are their own games, mostly against players near their level.
5. Winning positions: the biggest step is at the top
Clearly winning here means 85% winning chances or more, an advantage of about five pawns. From there, players at 600 still let 23% of those games slip to a draw or a loss, at 1000 20%, at 1200 and 1700 19%, at 2000 18%. The biggest drop comes at the top: 14% at 2500.
How different are a 1000 and a 1200?
A 1200 should score about 76% against a 1000. Yet on every measure here the two sit close together: 4.1 and 4.0 blunders a game, 1 lost game in 10 decided in the opening at both, 1.12 and 1.06 lead changes a game. On most measures the bigger steps come above 1200.
Why does a small edge win so many games?
In the research above, how hard a position was mattered more for blunders than rating: a modest step up in difficulty raised the blunder rate more than a 600-point rating difference lowered it. Stronger players also recognise more. Chess research estimates that masters know tens of thousands of patterns, yet look only slightly deeper ahead: about one extra half-move, one side’s turn, for every 400 points of rating (Chase and Simon, 1973; Gobet and Simon, 1996). Our numbers mostly compare players with opponents near their own level, so they can’t show how much harder a stronger opponent makes your positions.
What a rating can’t show you
A rating says how often you win. It can’t say which mistakes keep coming back in your games. Better at Chess reads your own Chess.com or Lichess games, names the mistakes you keep making, shows you how often each one happened, and builds your training from positions you played.
Get Better at Chess on the App Store. For iPhone.
All the numbers
Every 100-point Lichess band from about 500 up, shown on Chess.com’s scale. The levels named above are these rows: 600 is 500–650, 800 is 650–800, 1000 is 950–1100, 1200 is 1100–1250, 1700 is 1550–1750, 2000 is 1900–2050, 2500 is 2500 and up. Analysed blitz games, September 2026, Chess.com scale. Download all the numbers as a CSV file, free to reuse with a link to this page (CC BY 4.0).
| Rating | Blunders a game | Big blunders | Opening losses | Punished | Lead changes | Wins let slip |
|---|---|---|---|---|---|---|
| 500–650 | 4.3 | 1.26 | 15% | 60% | 1.29 | 23% |
| 650–800 | 4.2 | 1.19 | 13% | 61% | 1.22 | 22% |
| 800–950 | 4.2 | 1.14 | 12% | 61% | 1.17 | 21% |
| 950–1100 | 4.1 | 1.09 | 11% | 62% | 1.12 | 20% |
| 1100–1250 | 4.0 | 1.02 | 10% | 62% | 1.06 | 19% |
| 1250–1400 | 3.9 | 0.99 | 8% | 63% | 1.02 | 19% |
| 1400–1550 | 3.9 | 0.94 | 8% | 63% | 0.97 | 19% |
| 1550–1750 | 3.8 | 0.90 | 7% | 64% | 0.93 | 19% |
| 1750–1900 | 3.7 | 0.84 | 6% | 65% | 0.87 | 19% |
| 1900–2050 | 3.5 | 0.79 | 5% | 66% | 0.81 | 18% |
| 2050–2200 | 3.4 | 0.74 | 4% | 67% | 0.77 | 18% |
| 2200–2350 | 3.3 | 0.69 | 3% | 68% | 0.72 | 17% |
| 2350–2500 | 3.2 | 0.64 | 3% | 69% | 0.68 | 16% |
| 2500+ | 3.0 | 0.60 | 2% | 70% | 0.60 | 14% |
Questions
At what Elo do you stop blundering?
Not in the games we measured. In analysed Lichess blitz games from September 2026 with at least 20 moves by each side, players rated about 2500 on Chess.com’s scale still made 3.0 blunders per game, against 4.1 at 1000. What falls much faster is big blunders, losses decided in the opening and how often games swing.
How many blunders per game is normal in chess?
In analysed Lichess blitz games from September 2026 with at least 20 moves by each side, about 4 per player per game from 600 to 1200 on Chess.com’s scale, 3.8 at 1700, 3.5 at 2000 and 3.0 at 2500, counting a blunder as a move that throws away 10 points or more of the player’s winning chances.
What separates a 1200 from a 1700?
Blunder counts are close: 4.0 against 3.8 a game in analysed blitz games. The bigger differences: fewer of a 1700’s losses are decided in the opening (1 lost game in 15 against 1 in 10), a 1700 makes fewer big blunders (0.90 against 1.02 a game) and their games swing less (0.93 against 1.06 lead changes).
How good is a 1700 chess rating?
By the Elo formula, a 1700 scores about 95% against a 1200 and about 15% against a 2000. In analysed blitz games, about 1 in 15 of a 1700’s lost games was decided in the opening, against 1 in 10 at 1200 and 1 in 41 at 2500.
Is a 1200 much better than a 1000?
By rating, yes: a 1200 should score about 76% against a 1000. In analysed blitz games they play alike: 4.1 and 4.0 blunders a game, and 1 lost game in 10 decided in the opening at both.
Research we cite
- Ashton Anderson, Jon Kleinberg and Sendhil Mullainathan, “Assessing Human Error Against a Benchmark of Perfection”, KDD 2016. About 200 million games from the Free Internet Chess Server; a blunder there is a move that worsens the result in endgames with six pieces or fewer, checked against perfect play.
- Reid McIlroy-Young, Siddhartha Sen, Jon Kleinberg and Ashton Anderson, “Aligning Superhuman AI with Human Behavior: Chess as a Model System” (Maia), KDD 2020. Our blunder threshold comes from here.
- Herbert Simon and William Chase, “Skill in Chess”, American Scientist, 1973.
- Fernand Gobet and Herbert Simon, “The Roles of Recognition Processes and Look-Ahead Search in Time-Constrained Expert Problem Solving”, Psychological Science, 1996.
- FIDE, Rating Regulations, table 8.1.2, for expected scores by rating difference.
How we measured
- Games
- Every rated standard game played on Lichess in September 2026, from the Lichess open database (public domain): 88,673,670 games, after leaving out games with a bot, games that were abandoned, unterminated or ended by a rules infraction, correspondence games and games without both players’ ratings.
- Games with computer analysis
- Every number comes from the 8,868,664 of those games that carry Lichess’s computer evaluation: games someone asked Lichess to analyse. They aren’t a random sample: upsets are far more common among them than among all games. Most of them are between players of similar ratings. Counting only games between players of the same band changes big blunders per game and points per move by at most 0.04, and the shares of games without a big blunder, of blunders punished and of blunders given back by at most 1.5 points. Lead changes are counted only between players of the same band. Blitz only, rated 500 and up on Chess.com’s scale: 6,186,989 player-games for blunders and mistake size (games with at least 20 moves by each side), 3,391,627 lost games, 3,703,616 that reached a clearly winning position, 19,798,428 chances to punish a blunder, 1,733,556 games between two players of the same band.
- Blunder
- A move that lowers the player’s chances of winning by 10 percentage points or more, by Lichess’s own formula for turning an evaluation into winning chances. That’s the threshold of the Maia research project, and looser than Lichess’s “??”. A big blunder lowers them by 30 points or more. A level position is one where the player’s chances were 35% to 65% before the move. Points lost a move: the winning chances a move gave away, averaged over all of a player’s moves; a move that doesn’t lower them counts as zero.
- Where a loss was decided
- In games that ended in a win, the move from which the winner’s chances stayed at 70% or more to the end. Opening: moves 1 to 10. Endgame: after that, once the queens, rooks, bishops and knights of both sides add up to 24 or less (queen 9, rook 5, bishop and knight 3). Losses on time from a position that wasn’t lost count separately.
- Punished a blunder
- The player’s move right after an opponent’s blunder, made while the opponent still had 20% to 80% chances. Punished: afterwards, the player kept at least half of what the blunder gave them. Blundered back: the reply was itself a blunder.
- Lead changes
- In games between two players of the same band: White leads at 65% winning chances or more, Black at 35% or less, and otherwise the lead stays where it was. A lead change is the lead passing to the other side.
- Wins let slip
- Games where the player had 85% winning chances or more (about +4.7) after one of their own moves, from move 10 on, and then lost or drew.
- Rating bands
- The player’s rating before the game, in Lichess’s 100-point bands, shown as their range on Chess.com’s scale. Each level on this page is the band its Chess.com rating falls in: 600 is 500–650, 800 is 650–800, 1000 is 950–1100, 1200 is 1100–1250, 1700 is 1550–1750, 2000 is 1900–2050, 2500 is 2500 and up. The tables show every band from 500 up; lower bands are left out because the conversion starts there. The ranges are rounded to the nearest 50, which moves an edge by at most 25 points, well inside the conversion’s spread.
- Expected scores
- The Elo formula, 1 / (1 + 10−D/400) for a rating difference D, which FIDE’s table follows and Chess.com’s and Lichess’s systems build on.
- Chess.com scale
- We converted each Lichess blitz rating to Chess.com’s scale with a comparison of players who have established blitz ratings on both sites: a Lichess 1555 matches a Chess.com 1200. Single players scatter around these matches: the middle half of players are within about 70 points of them. As a rough check, we compared how often players on each site make errors in comparable positions in rapid: at Chess.com 1200 and 1600, the Lichess ratings with the same error rates came within about 70 points of the comparison’s, and each of our estimates is only good to about 100 points either way. Matching error rates support the conversion without proving it.
- Rounding
- Whole percentages, and one or two decimals for counts per game, because the conversion is approximate.
- Privacy
- We kept no usernames, game links or game IDs. Every published number comes from groups of at least 1,000 player-games, games or chances.
- Fixed first
- The definitions of blunders, where a game was decided and winning positions were written down before we looked at any result. Mistake size, punishing and lead changes were defined on 10 November 2026, after the first results and before these were computed. The conversion to Chess.com’s scale was added afterwards, and the 100-point bands on the same day, before any result at that size was seen.