All the gambits ranked by Lichess Explorer statistics

Intro

My browser extension LiChess Tools uses some static lists to display openings and gambits, even show which moves lead to gambits in analysis mode. if I have then a list of gambits and also access to the Lichess Explorer database, wouldn't it be trivial to find the most winning gambits?

So I got the info I needed, started that marvel of mathematical precision and statistical prowess, Excel, and went on to find out which is the best gambit of them all!

And I learned that opening databases are a mess and I don't know what a gambit is supposed to be!

Methodology

First of all, the data: for each gambit we have the following information:

• gambit name
• gambit color - which side plays it
• wins - number of games won by the gambit color at the final position of the gambit move list
• draws
• losses
• average ELO rating
• PGN
• FEN - of the final position in the gambit move list

From these we can also infer:

• the total number of games played
• the win ratio - wins/total
• the ELO win ratio - wins/total * average ELO

We then need to clean the data:

• some gambits are reached by multiple paths - we only keep one row per FEN
• some gambits are never played - we only keep the ones that have been played at least 100 times
• some gambits are losing or drawish - we only keep the ones that have a win ratio of 50% or more
• replace the non-ASCII letters with their ASCII counterparts

From a total of 867 gambits for both Black and White, 339 remained.

Sort by win ratio

I started with sorting by win ratio descending. The best gambit of them all is... the Landau gambit for White?!
Coming against the Caro Kann, it goes like this: 1.e4 c6 2.d4 d5 3.Bd3 Nf6 4.e5 Nfd7 5.e6 . It's the main line Caro, followed by a strange sixth most played move 3. Bd3, but which has no name, after which Black plays 3...Nf6? Just whose gambit is this? From a comfortable equality the eval jumps to +1.5, from where all the moves are the best moves in the position.

This "gambit" has been played 2987 times from which 87% were wins for White. The average ELO is 1714 and the "win ELO" is 1496. This first candidate for GOAT (gambit of all times) disappoints. It shows us that the database calls gambits moves that may not gambit anything.

The next three rows are similar gambits: Tennison Gambit, coming from Scandi, Caro and French. They all score around 80% for White! Igor Smirnov was on to something. In total, they have been played 741386 times and the average ELO is also around 1700. The win ELO is around 1400.

Let's look at the line coming from the Scandinavian Defense, which is the most played by far: 1.e4 d5 2.Nf3 dxe4 3.Ng5 Nf6 4.d3 exd3 5.Bxd3 h6 6.Nxf7 . Normal Scandi, followed by a weird Nf3 (the Reti knight), Black takes, knight moves to g5 (the Abonyi Gambit). Nf6, d3, pawn takes, bishop takes. At this moment the eval is -0.8. Here Black MUST play Nc6 or e6, maybe even Bg4. Instead, Black plays h6! leading to an eval of +3! Nxf7 is the best move in the position!

Perhaps the gambit should end at Bxd3. Black's h6 is falling into a trap and White's Nxf7 is the trap springing shut. While the win ratio in this position is still a decent 58%, it is far from 87%! Yet, without taking on f7, is it still a Tennison's? And why is the most significant move called the Abonyi Gambit, then?

The next one is the Zombie King Gambit, against the Philidor Defense! Similar to Tennison, the line ends with Bxf7 after one more greedy Black capture jumps the eval from +0.5 to +2.2. Funny, though, the initial low depth evaluation is close to +1: 1. e4 e5 2. Bc4 d6 3. Nf3 h6 4. d4 exd4 5. c3 dxc3 6. Bxf7+

We know that gambits have structure: you propose the gambit, the opponent has a chance to accept or reject it. Should gambits, then, be defined by at least three lines? The gambit proposal: up to 5. c3, the accepted line: 5...dxe3 and anything else is a rejected line? What about the actual killer move? Without it, it would be like a joke without a punchline!

The Zombie King Gambit has been played 4735 times and has a win ration of 73%. The average ELO is 1600 and the win ELO is 1160. However, if we use the convention above, the proposal position has been reached 141678 times, with a decent win ratio of 61% and the average ELO of the people playing the accepted line (92000 games) is 1657.

Let's end the exploration of the Win ratio sorting with the ... Damiano Gambit. This is ridiculous! The Damiano defense is a really really really bad response to 1.e4: 1. e4 e5 2. Nf3 f6 3. Nxe5 fxe5 4. Qh5+ g6 5. Qxe5+ Qe7 6. Qxh8. After the disastrous 2...f6, Black doubles down with the terrible 3... fxe5, where the only options for Black are checkmate or losing a lot of material. This is the Damiano Defense. The simplest line is losing the rook. Do you want to know what the Damiano Gambit is? Taking the rook.

That's right. This "gambit for White" ends after the opponent blunders a rook and White takes it. Maybe the gambit is 3. Nxe5? But that's the best move in the position. 900000 games have reached that position and 700000 times Black takes the knight, losing the game. But is it still a gambit if the opponent has to blunder in order to get to the starting position and we play the best move in the position?

The average ELO for this is 1300 and the win ELO is 900 or so. The Damiano Gambit, an abomination.

The win ELO

OK, let's sort by this meta number that we conjured by multiplying the win ratio with the average ELO. If the win is high and the people playing it are good, then surely it is a measure of gambit quality.

With this ordering, the Landau and the Tennison are still top, but some other less played gambits have risen close:

• the Yu Gambit - from the Ruy Lopez against the Berlin Defense - 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.Re1 Nd6 6.Nxe5 Be7 7.Bf1 Nxe5 8.Rxe5 O-O 9.d4 Bf6 10.Bd3
• the Matsukevich Gambit - against the Alekhine Defense - 1.e4 Nf6 2.e5 Nd5 3.c4 Nb6 4.c5 Nd5 5.Nc3 Nxc3 6.dxc3 d6 7.Bg5
• the Petrosian Gambit - from the Torre Attack against the Classical Defense - 1.d4 Nf6 2.Nf3 e6 3.Bg5 c5 4.e3 b6 5.d5

All of these gambits have been played less than 5000 times, but the average ELO for them is around 2200! The win ratio for these gambits is above 60%.
Honorable mentions: the Prianishenmo Gambit and Peresypkin's Sacrifice. I mean, the names alone!

The losers

So if these are the top gambits, what are the really bad ones, the ones that always lose?

• the Dudweiler Gambit for White - 1.f4 d5 2.g4
• the Miguel Gambi for White - 1.e4 e5 2.Bb5 Bc5 3.b4
• the Coca-Cola Gambit for White - 1.g4 g5 2.f4
• the Fontaine Countergambit for Black - 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.b4 b5
• the Zilbermints Gambit for Black - 1.Nf3 d5 2.c4 b5 - oh, Lev!

All these gambits manage a measly 30% win ratio, with ELOs around 1300.

Conclusion

This experiment created more questions than answers. What is a gambit? Where does it start and where does it end? What is the best way to clean the gambit list to get the maximum information and benefit?

You can play with the same data by downloading the gambits list.

Enjoy!