Latvian Gambit: Is It Sound, and Why It Wins Games Anyway

The Latvian Gambit

You hang a pawn on move two. Six moves later White's knight is eating a rook in the corner — and it never gets back to defend the mate.

1.e4 e5 2.Nf3 f5 is the King's Gambit played backwards, a move down. Stockfish has White at +2.2 the moment you play it, and you should know that before you touch the f-pawn. What the engine cannot price is the shape of the position: White's most natural, most human idea is to send his knight raiding — e5, f7, h8 — and by the time it has swallowed your rook it is a knight in a corner with no route home. That is the whole gambit. You are not trading a rook for a pawn. You are trading a rook for the piece that was supposed to defend his king.

The line, move by move

1. e4 e5 2. Nf3 f5 3. Bc4 fxe4 4. Nxe5 Nf6 5. Nf7 Qe7 6. Nxh8 d5 7. Bb3 Bg4 8. f3 exf3 9. Kf2 Ne4+ 10. Kg1 f2+ 11. Kf1 Bxd1

1. e4 (opponent) King's pawn. The most-played first move there is.
1... e5 (you) Mirror it. Still the most normal position in chess.
2. Nf3 (opponent) Nf3 attacks your e5 pawn. Every beginner book says defend it with ...Nc6. Ignore every beginner book.
2... f5 (you) f5!! Your e5 pawn is already hanging and you just put a second one en prise. Know what you are doing: depth-18 Stockfish calls this +2.2 for White. It is not sound. It is a bet that he cannot play twelve accurate moves in a row — and across 3.8 million games at 1000-1600, he usually can't.
3. Bc4 (opponent) Bc4 — 6% of the time. He develops at f7 and leaves both centre pawns alone. This is the rarest of his five tries and by far the most dangerous for him.
3... fxe4 (you) fxe4! Grab it — 30% of players do. Your f-pawn is gone from f5, which means the f7 square is now an empty hole, and White can smell it.
4. Nxe5 (opponent) 91% take on e5. Of course they do — the pawn was hanging for two moves. His knight is now sitting on e5 looking at f7, d7 and g6 all at once.
4... Nf6 (you) Nf6 — 11% play it. The popular move is 4...d5 (53%), which is level and safe (+0.3 after 5.Qh5+). This one is different: it leaves d8 and h8 on the same knight-fork colour and says, out loud, come and take my rook.
5. Nf7 (opponent) Nf7! — 68% of the board plays this, and who could blame them: the knight hits your queen AND your rook from one square. You are about to lose the exchange on purpose.
5... Qe7 (you) Qe7 — 92% of players find this, and it is right. Move the queen, leave the rook. The queen lands on e7 where it eyes e-file and h4 at the same time; the rook on h8 was doing nothing anyway.
6. Nxh8 (opponent) Nxh8 — 98%. Almost nobody declines. Look hard at that knight: it is on h8, hemmed in by your g7 and h7 pawns, and there is no square in the world it can reach that matters. You are down a rook and up a piece that has left the game.
6... d5 (you) d5! — 87% of players find it. Hit the bishop, free your light-squared bishop, and start counting: for the rest of this line White is effectively playing without a knight.
7. Bb3 (opponent) Bb3 — 57%, the natural retreat that keeps the bishop on the a2-g8 diagonal. And it is the move that loses: the engine goes from +2.5 White to -5.1 in one bishop step.
7... Bg4 (you) Bg4!! — 61% play it. The bishop lands on g4 staring straight down the diagonal at his queen on d1. Only f3 and e2 stand between them, and both are empty. He has to deal with this right now, and every way of dealing with it is a disaster.
8. f3 (opponent) f3 — 86%. The most natural move on the board: interpose a pawn, threaten the bishop, get on with life. It also just tore the roof off his own king, and his defending knight is in the corner on h8.
8... exf3 (you) exf3! Take it. Your pawn is now on f3, one square from his king, attacked by nothing that matters. From here the engine reads mate in 14 no matter what he does.
9. Kf2 (opponent) Kf2 — 90%. He steps toward the pawn to eat it. The other 9% play Kf1 and lose the same way. Nobody has time to castle: he never will in this game.
9... Ne4+ (you) Ne4+! — 69% of players find it. Check from e4, and notice what that knight now guards: f2, d2, c3, g3. Every escape hatch on his king's side of the board runs through a square this knight covers.
10. Kg1 (opponent) Kg1 — 61%, the instinct to hide in the corner. It is the most popular and it loses the fastest.
10... f2+ (you) f2+! — 82% of players push it, and it is right. The pawn steps to f2 with check, and here is the quiet part: it just cleared f3, so your bishop on g4 now has a completely open road to d1.
11. Kf1 (opponent) Kf1 — 100% of 642 games, because it is the only legal square on the board. Your knight on e4 covers d2 and g3; the pawn on f2 covers g1; the queen on d1 is his own.
11... Bxd1 (you) Bxd1!! Bishop takes queen. This is why f2+ went in first — the pawn vacated f3 and the diagonal g4-f3-e2-d1 opened like a door. He is a queen down, his king is on f1, and his knight is still on h8.

Every branch cashes out

9 coached branches in this family. Engine-checked finishes:

Bc4 / Nf7 / f3 / Kg1+9.0 by force
Queen taken on move eleven, and Stockfish reads mate in seven from here. Trace it back: he won a rook on move six and every single move after that was reasonable. The rook was the bait. The knight in the corner was the price.
Bc4 / Nf7 / f3 / Ke3Forced mate
Mate on the board, chess.js-verified, delivered by a pawn that moved once. His king walked from e1 to d4 in five moves because every check pushed it forward — and the knight that could have covered c6, b4 and d5 was on h8 eating a rook.
Bc4 / Nf7 / Qxg4+4.9 by force
Queen for bishop, +4.9 to you, on move eight. He is a queen and a functioning knight short with an uncastled king — and 86% of the field blocks with f3 instead and gets mated. There is no good version of this position for White.
Bc4 / Bf7++5.1 by force
+5.1 to you. He found the strongest move in the whole tree on move five and still lost a piece by move eight, because a bishop on f7 with a black king on e7 is not an attacker — it is a hostage. Same lesson as the knight on h8, one file over.
exf5Sound line
+0.3 for White — a rounding error, in a line where you were "objectively lost" nine moves ago. The extra pawn on f5 is coming back to the bishop, you have both bishops and an open d-file, and White has one piece developed. This is what the Latvian is actually for.
Nxe5Sound line
+1.5 for White, which is the honest price of the main line — he plays it right, he stays better. But he is BETTER, not winning, and this is the position 73% of the field steers into. Every pawn is back and you have the structure you wanted. Some days the gambit just gives you a game.
Nc3 / Nc3Sound line
+0.4 for White — statistically nothing, from a gambit the engine hated at move two. He played the whole line correctly and got an equal game out of it. Against most people, that already feels like losing.
Nc3 / Neg5+4.0 by force
A free knight on move six, +4.0 to you. He moved one knight to safety and left the other one hanging on a diagonal that opens the instant Black plays ...e5 — which was move one. Look at loose pieces before you look at threats.
d3Sound line
Dead level — 0.0, measured. You have both bishops, an active queen and a half-open f-file, from an opening Stockfish scored at +2.2 for him seven moves ago. Nothing was won here. Everything was clawed back, and that is the honest, everyday result of the Latvian.

Why it works

The Latvian is a bad bet made on purpose. Stockfish scores 2...f5 at plus 2.2 for White the moment it lands, and you should know that before you touch the f-pawn. What the engine cannot price is what White's most natural plan does to his own position: the knight raid. It goes e5, then f7, then h8, and by the time it has swallowed your rook it is a knight in a corner, hemmed in by your g7 and h7 pawns, with no route home that matters. You are not trading a rook for a pawn. You are trading a rook for the piece that was supposed to defend his king.

The Buried Knight branch shows the price he pays. 68 percent fork on f7, 98 percent take the rook, and then one natural bishop retreat, Bb3, flips the eval from plus 2.5 for White to minus 5.1 in a single move. From there the punishment is measured to the end: the f3 block that 86 percent play tears the roof off his own king and the engine reads mate in 14, finishing with the bishop taking the queen on d1 down a diagonal your f2 pawn push opened like a door. The king that walks forward instead gets mated on move thirteen by a pawn from c7, chess.js-verified, protected by a bishop that never left its starting square.

Even White's best try is not a safe one. 5.Bf7+ is the strongest move in the whole tree, found by 15 percent and worth plus 3.1, and it still ends in tears when he gets greedy: the checking bishop becomes a hostage the moment your king attacks it and the e5 knight is pinned to defending it, and the branch closes with Qxd1+, dissolving his pin and leaving you plus 5.1. Same lesson as the knight on h8, one file over.

How to punish the Latvian Gambit

White refutes the Latvian by refusing the raid. The principled main line is 3.Nxe5, and played correctly it simply keeps plus 1.5: back the knight up with d4, retreat it to c4 when the d-pawn kicks it, and accept that better is not winning. In the Bc4 line the move is Bf7+ before touching the rook, worth plus 3.1, followed by the timely retreat 6.Bb3, which keeps plus 2.9. The greedy d4 prop that tries to hold f7 forever is where the hostage-taking starts.

What the data warns against is exactly what feels best: the rook on h8. It costs a knight's entire career. And the most common move of all, 3.exf5, is quietly deflating: 44 percent play it, Black kicks the knight with e4, the c8 bishop collects the f5 pawn back, and the game lands at plus 0.3, a rounding error from a position that was supposed to be winning.

How to avoid falling into it

Four pieces of folklore died during the authoring of this tree, and every one is a move you will be tempted to play as Black. ...Qxf5 while the knight sits on d4 loses the queen outright to Nxf5, measured at plus 14.7 for White. The ...Bxf5 pawn pickup in the same structure drops a whole bishop, plus 11.4, because the recapturing bishop on d3 covers f5. And the greedy ...Nxe4 grab in the quiet d3 line walks into Bxe5, forking the h8 rook at plus 3.6. All three look obvious over the board. That is exactly why they were measured instead of trusted.

Also know which door the famous rook offer lives behind: the Nf7 fork only exists after 4...Nf6, which is why this tree coaches that move. If you want the level, safe route instead, 4...d5 is what 53 percent of players choose, and it stands plus 0.3 after the queen check. Take the dare on purpose or do not take it at all.

Questions people actually ask

Is the Latvian Gambit sound?: No. Depth-18 Stockfish has White at plus 2.2 the moment 2...f5 lands, and this walkthrough says so before you play a single move. It is a bet that White cannot find twelve accurate moves in a row, and across 3.8 million games at 1000-1600 he usually cannot.
What is the refutation of the Latvian Gambit?: The principled 3.Nxe5, met with ...Qf6 and backed by d4, keeps White at plus 1.5 with best play: better, but only better. The sharpest single move in the tree is 5.Bf7+ in the Bc4 line at plus 3.1. Meanwhile the most common choice, 3.exf5, lets Black claw all the way back to plus 0.3.
Why does Black give up the rook on h8 in the Latvian?: Because the knight that takes it never comes back. On h8 it is fenced in by the g7 and h7 pawns, so White is effectively playing a piece down everywhere that matters, and his king has lost its best defender. In the main punished line, one natural retreat later, the eval swings from plus 2.5 White to minus 5.1 and the engine starts reading mate.
What happens if White just takes the f5 pawn?: That is the most common answer, 44 percent, and the calmest for you. Push e4 to kick the knight instead of chasing the pawn, aim the queen down the f-file, challenge the d4 knight with ...Nc6, and the gambit pawn comes home to the bishop. The measured result is plus 0.3 for White: statistically nothing.
At what rating does the Latvian Gambit work?: Every frequency in this walkthrough comes from blitz and rapid games at 1000-1600, where 68 percent of White players take the f7 fork bait and 86 percent block the bishop attack with the losing f3. Against stronger opposition you meet the plus 1.5 and plus 3.1 lines more often, which is the honest ceiling of the opening.

Drill the Latvian Gambit against a real opponent

The trainer above plays every branch with a coach explaining each move. When you are ready, take it into a live game, or paste your Chess.com or Lichess username and see which openings actually cost you rating.

Play it for real Get your free scout report

More opening weapons

New to openings? Learn openings without memorizing explains the approach behind every walkthrough here, and the free AI chess coach guide covers what the rest of the app does without a card.