An atomic game is usually decided by an explosion, not a mate. We read 294,056 rated Lichess atomic games from October 1 to 29, 2026 and counted 230,424 of them: 55.4% ended with a king blown up and 13.2% with ordinary checkmate. Below: the rules, where to play, then the data on how games end by rating, how fast, what a capture costs, which openings people choose, and how atomic blunders compare with blitz.
The short answer
Atomic chess is chess where every capture detonates, and you win by blowing up the enemy king. It is two player and single board. In our sample 55.4% of games ended in an exploded king, and that share falls from about 80% under 1000 to about 26% at 2200 and above. White scored 0.569 points per game.
The rules of atomic chess
Atomic is standard chess with one new rule about captures. Everything below is attributed to Wikipedia, Chess.com's support pages and the XBoard rules page, with Lichess's behaviour noted where we could see it second hand.
- A capture explodes. The capturing piece, the captured piece and every non-pawn piece on the eight squares around the capture are removed, both colours. Wikipedia's wording is that all surrounding pieces of both colors other than pawns are removed from play.
- Pawns next to a blast survive. A pawn that is captured, or that is itself the capturing piece, is removed. Chess.com says pawns are immune to explosions, which is loose: they are immune as bystanders only.
- You win by exploding the enemy king. Chess.com's support page says the game ends if a king is destroyed or checkmated. If one blast would remove both kings, the capture is illegal.
- A king can never capture. The capturing piece explodes, so it would take its own king with it. Wikipedia: captures that result in the explosion of a player's own king are illegal.
- Kings may touch. Wikipedia and Chess.com both say kings can stand on adjacent squares without check, since capturing the neighbour would blow up the capturer's own king too. So a king standing next to the enemy king is safe from explosions, because any capture beside it would also blow up the attacker's own king. That last point is our inference from the rule.
- Castling and promotion follow standard chess. Only XBoard spells this out: a pawn promotes to a queen, rook, bishop or knight, never a king. Stalemate is a draw.
Check, as Lichess and Chess.com handle it
Sites differ here. The XBoard description is destroy the king, with check mattering mainly for castling, and per Wikipedia the Internet Chess Club version does not enforce check at all. Lichess and Chess.com do enforce it: a move is illegal if your king is in check after the explosion, and checkmate also wins. An exploding capture can still win while you are in check, as long as it removes the enemy king without removing yours. We read the Lichess wording second hand, so check the rules page of the site where you play for edge cases.
Where it comes from
Per Wikipedia, atomic appeared in 1995 on the German Internet Chess Server, from rules a user collected from friends who played offline. No inventor is named. It reached the Internet Chess Club in 2000, FICS in 2003, Lichess in 2015 and Chess.com in late 2020.
Where to play atomic chess online
As of November 2026, Lichess and Chess.com both offer it, and Wikipedia also lists the Internet Chess Club and FICS. On Lichess atomic has its own rating, separate from standard chess, so treat your first atomic number as a new scale. Chessdrive is not an atomic server, so for rated games you want one of them. Our guide to playing chess online for free compares the main sites for standard chess. For other variants see our crazyhouse chess guide and Chess960 and its shuffled start.
Is White winning?
Wikipedia says White's first move advantage is much greater than in standard chess, and also that no attempt to prove a win for White has succeeded. Engines and players believe White is much better, but that is unproven. Our games are human games, so they show how White scores in practice, not what perfect play gives.
How 230,424 atomic games end
We streamed Lichess's open atomic database (CC0). The October 2026 file lists 324,630 games; our stream read 294,056 of them (90.6%) before it closed on October 29, so the data covers October 1 to 29, 2026. Of those, 230,424 counted after dropping 4,643 BOT games and 58,989 games where the two ratings differed by more than 400. Bands use the average of the two ratings. They are Lichess atomic ratings, a separate pool from standard ratings, so a 1500 here is not a 1500 in blitz. Every table below uses all speeds unless it says otherwise.
| Average rating | Games | Ended by king explosion | Won on time | Average length (plies) | White's points per game |
|---|---|---|---|---|---|
| Under 1000 | 17,644 | 80% | 4% | 13.7 | 0.559 |
| 1000 to 1399 | 71,290 | 69% | 7% | 19.0 | 0.546 |
| 1400 to 1799 | 84,268 | 55% | 12% | 28.0 | 0.573 |
| 1800 to 2199 | 49,567 | 32% | 21% | 43.3 | 0.596 |
| 2200 and above | 7,655 | 26% | 30% | 48.1 | 0.594 |
| All games | 230,424 | 55% | 12% | 28.1 | 0.569 |
Source: Lichess open database, October 2026 atomic file, CC0, all speeds. A ply is one move by one side. "Won on time" is the share of games ending on the clock. Percentages in the table are rounded to whole numbers. Draws were 2.0% of games.
What the table says
- The explosion share falls steeply with rating. From about 80% of games under 1000 to about 26% at 2200 and above. Ordinary checkmate moves the other way, from roughly 7% under 1000 to 15.6% at 1800 to 2199, and it is 13.9% at the top band.
- The clock matters more as ratings rise. About 30% of games at 2200 and above end on time, against about 4% under 1000. That mixes in speed, since 6,785 of the 7,655 games at 2200 and above are bullet, so read it as a rough guide.
By speed
Bullet, blitz, rapid and classical were 105,984, 97,406, 24,556 and 2,478 games. The king explosion share was 49.1% in bullet, 75.6% in rapid and 75.1% in classical, and about 57% in blitz. Bullet games end on time far more often: 21.3% against 4.9% in blitz and 1.8% in rapid. Within the 1400 to 1799 band the explosion share is 52.9% in bullet, 53.4% in blitz and 73.7% in rapid, so slower games end in explosions more even at a matched rating.
How fast games end on the board
Of the 230,424 games, 158,088 ended on the board, meaning a king explosion or a checkmate. The next numbers are shares of those games, not of all games. Ply 20 is the end of move 10 and ply 40 is the end of move 20. Across all bands 51.3% of the on board endings came before ply 20 and 81.5% before ply 40, and the average on board ending came at ply 25.0.
| Average rating | Games ended on the board | Ended before ply 20 |
|---|---|---|
| Under 1000 | 15,308 | 81.2% |
| 1000 to 1399 | 58,221 | 64.3% |
| 1400 to 1799 | 57,708 | 47.4% |
| 1800 to 2199 | 23,800 | 15.7% |
| 2200 and above | 3,051 | 7.4% |
Source: same file, all speeds, shares of games ended by explosion or checkmate. At 2200 and above, 41.6% of those endings came before ply 40, against about 97% under 1000. Below 1400, most on board endings come before move ten.
White's score
White scored 0.569 points per game overall, where a win is 1 and a draw is half. By band it was 0.559 under 1000, 0.546 at 1000 to 1399, 0.573, 0.596 and 0.594 in the three bands above. White also took 56.5% of the games decided by an explosion or a mate, and that share is 61.4% at 2200 and above. A 0.569 average is not evidence of a forced win, and we do not compare it with standard chess.
What a capture costs
Every capture is a small bomb, so we replayed every tenth game on the board (23,077 games, 86,724 captures, no parse failures) and measured what each one removed. Across 6,471,434 moves, 13.4% were captures. Capture rate is 12.8% to 14.5% of moves in the four bands from 1000 up.
The size of a blast includes the capturer and the captured unit, so the overall average of 2.69 units removed per capture means about 0.7 extra units beyond those two. We also counted two things: how often a capture costs the mover its own queen, which is usually because the queen itself made the capture, and how often the capture removes only the target and the capturer, a clean trade.
| Mover's rating | Units removed per capture | Capture costs the mover its queen | Clean trade, only the two units |
|---|---|---|---|
| Under 1000 | 3.0 | 24.1% | 47.1% |
| 1000 to 1399 | 2.9 | 20.3% | 48.5% |
| 1400 to 1799 | 2.7 | 15.9% | 55.2% |
| 1800 to 2199 | 2.5 | 12.9% | 62.9% |
| 2200 and above | 2.4 | 10.4% | 66.6% |
| All | 2.7 | 16.0% | 56.4% |
Source: same file, replayed captures only, by the mover's own rating. Averages are rounded to one decimal here.
Both trends run the same way. Weaker players set off bigger blasts, and in 24.1% of their captures the mover loses its own queen, against 10.4% at 2200 and above. Strong players mostly trade cleanly: two thirds of their captures remove only the two units involved. Counts alone cannot say whether those queens were sacrificed on purpose.
Do atomic games produce more blunders?
Only games that Lichess analysed carry evaluations. In this file that is 19,558 games, a self selected subset of the 230,424, since people who request analysis are not a random sample. Following our blitz and bullet articles, a blunder is a move from ply 10 onward that drops the mover's win probability by 30 points or more. The atomic evaluations come from Lichess server analysis with a variant engine.
Across all speeds, blunders per 100 moves were 12.3 at 1000 to 1399, 8.3 at 1400 to 1799, 6.0 at 1800 to 2199 and 4.8 at 2200 and above, and 8.1 for all moves together. Under 1000 the figure was 16.0, but it rests on a small sample. The comparison below uses blitz only, against the standard blitz data from our blitz chess analysis, with the same 30 point rule and the same ply 10 start, matched by the mover's band.
| Mover's rating | Atomic blitz, blunders per 100 | Standard blitz, blunders per 100 | Ratio |
|---|---|---|---|
| Under 1000 | 14.7 (small sample) | 5.7 | 2.6 |
| 1000 to 1399 | 11.8 | 4.4 | 2.7 |
| 1400 to 1799 | 7.8 | 3.4 | 2.3 |
| 1800 to 2199 | 5.2 | 2.6 | 2.0 |
| 2200 and above | 3.8 (small sample) | 1.9 | 2.1 |
Source: atomic from the October 2026 atomic file; standard from our Lichess blitz data of 2026-11-14. Ratios divide the unrounded rates. Two atomic rows rest on small samples and are marked.
The direction is clear in the three rows with solid samples: atomic blitz runs at about 2.0 to 2.7 times the standard blitz rate for the same band. Two caveats apply. A fixed 30 point rule catches more moves where one capture can swing an evaluation from equal to lost, so this is not 2.5 times as many bad moves as a human would judge. And only analysed games count.
Which openings people play
Across 228,522 first moves, 1.Nf3 was played 50.5% of the time, ahead of 1.e3 at 15.6% and 1.e4 at 15.6%, and 1.Nh3 at 6.4%. The choice shifts sharply with rating.
| Average rating | 1.Nf3 | 1.e4 | 1.Nh3 |
|---|---|---|---|
| Under 1000 | 36.6% | 34.7% | under 1% |
| 1000 to 1399 | 48.6% | 22.2% | 1.5% |
| 1400 to 1799 | 49.4% | 13.4% | 6.0% |
| 1800 to 2199 | 58% | 5.1% | 14.1% |
| 2200 and above | 60.8% | 2.3% | 20.1% |
Source: same file, first moves by the average rating of the two players. At 2200 and above 1.e3 takes about another 10%. Our counts cannot say whether 1.Nh3 is any good.
Why 1.Nf3
The knight eyes e5 and g5, and from either square Nxf7 is a threat: f7 sits next to the black king on e8, so the blast takes the king with it and the game ends. The reply 1...f6 covers both squares. Per Wikipedia many atomic openings aim to explode Black's d, e or f pawn, and it describes a quick loss after 1.Nf3 d5 2.Ne5. Black is commonly said to need 1...f6, though sources differ on whether other replies hold, so treat that as community opinion rather than engine fact. The data agrees with it in one way: the five most common four ply lines overall all start 1.Nf3 f6, and together they are about 22% of games.
| First four plies | Games | Share |
|---|---|---|
| 1.Nf3 f6 2.Nc3 c6 | 16,648 | 7.4% |
| 1.Nf3 f6 2.Nd4 Nh6 | 10,025 | 4.5% |
| 1.Nf3 f6 2.Nc3 Nh6 | 8,584 | 3.8% |
| 1.Nf3 f6 2.e3 d5 | 6,989 | 3.1% |
| 1.Nf3 f6 2.e3 e6 | 6,624 | 3.0% |
| 1.Nf3 e5 2.Ng5 f5 | 5,860 | 2.6% |
Source: same file, all bands, 224,445 games with four plies. At 2200 and above the leaders are 1.Nf3 f6 2.Nc3 Nh6 (10.7%) and 1.Nf3 f6 2.Nc3 c6 (10.7%), with 1.Nf3 e5 2.Ng5 f5 at 6.1%. Under 1000 the single most common four ply line is 1.Nf3 e5 2.Ng5 f6 at 2.8%, and 1.e4 d5 2.exd5 Qxd2# is 1.8% of games: the capture on d5 removes both pawns and opens the d-file, and the queen then takes on d2 beside the king on e1. If you want a plan rather than a list, our article on learning plans instead of memorizing openings applies with extra force, because here a single capture next to your king ends the game.
What to take from the numbers
These are our readings, not engine verdicts. Below 1400, most games that end on the board end before move ten, so checking each capture next to your own king before you move is worth more than any opening line. From 1800 up, games turn into clean trades and the clock plays a bigger part. For how time pressure alone changes games, see our bullet chess data.
Method and sources
Data: Lichess open database, rated atomic games, October 2026 file (CC0), streamed with no local copy, covering October 1 to 29, 2026 because the stream closed early. We read 294,056 of the 324,630 games in the dump and counted 230,424 after dropping 4,643 BOT games and 58,989 games with a rating gap over 400. Bands use the average of the two ratings for game tables and the mover's own rating for capture and blunder tables. Speed is Lichess's estimated duration, base plus 40 times the increment. An ending is a king explosion if the last move is a capture with a checkmate sign, checkmate if it has the sign and is not a capture, a time forfeit if the termination says so, and anything else is resignation, abandonment or a draw. That rule agreed with the board on 15,825 of 15,828 replayed endings. Explosion sizes come from every tenth game replayed on python-chess's atomic board (23,077 games). A blunder is a move from ply 10 that drops the mover's win probability by 30 points or more, on the 19,558 games carrying Lichess evaluations. Standard comparison: our Lichess blitz blunder table of 2026-11-14, same rule. Rules, history and the White advantage statement: Wikipedia's atomic chess article, Chess.com's terms and support pages, and the XBoard rules page. We could not read Lichess's own variant page and leave out the en passant blast rule, which we could not verify. Facts are current as of November 22, 2026.
Frequently asked
- What is atomic chess?
- Atomic chess is chess in which every capture causes an explosion. The capturing piece, the captured piece and every non-pawn piece on the eight squares around the capture are removed, whatever their colour. You win by exploding the enemy king or by checkmate.
- Do pawns explode in atomic chess?
- Pawns next to a blast survive it. A pawn that is itself captured, or that makes the capture, is removed like any capturing piece. So pawns are protected from collateral damage, not from being taken.
- Can a king capture in atomic chess?
- No. A capturing piece explodes, so a king that captured would remove itself, and any capture that blows up your own king is illegal. For the same reason kings may stand on neighbouring squares without it counting as check.
- Is White winning in atomic chess?
- Players and engines believe White's first move advantage is much larger than in standard chess, but nobody has proved a forced win and the game is not solved. In our Lichess sample White scored 0.569 points per game across all ratings and speeds.
- What is the best first move in atomic chess?
- 1.Nf3 is the classic choice and it was played in 50.5% of the 228,522 games we counted, rising from 36.6% under 1000 to 60.8% at 2200 and above. Its idea is Ne5 or Ng5 followed by Nxf7, which explodes the area around the black king.
- Where can you play atomic chess online?
- Lichess has offered it since 2015 and Chess.com since late 2020, according to Wikipedia. The Internet Chess Club and FICS also carry it. Atomic has its own rating on Lichess, separate from standard chess.
Play atomic for the explosions, review your standard games for progress
Chessdrive is not an atomic chess server, so play the variant on Lichess or Chess.com. For the standard games you play every week, a Stockfish review in Chessdrive checks every move and explains the ones that mattered at your level. On iOS and Android, $5.99/month or $39.99/year on the US storefront with a free trial. Prices vary by region.
Keep reading
Updated