Playing10 min read

King's Gambit: What Each Rating Actually Plays

Fischer said the King's Gambit loses by force. Chessdrive's Maia3 and Stockfish data shows what each rating band plays after 2.f4, where the engine's favourite replies go unfound, why 4.Bc4 is the human choice and 4.d4 the engine one, and which line a club player should pick.

The starting position

Bobby Fischer wrote that the King's Gambit, 1.e4 e5 2.f4, loses by force, and club players have kept playing it for decades anyway. We ran the maia3-5m checkpoint (how Maia chess actually works) on six positions along the King's Gambit at seven rating bands from 1000 to 2200, and scored every legal move with Stockfish 18 at depth 20 (what the Stockfish chess engine is). This is the same method as our chess openings by rating dataset. Maia3 is a model that predicts which move a player of a given rating is likely to choose. It is not a database of real games. We tried to cross-check against real games from the Lichess explorer and access was refused, so every percentage below is a model prediction, not a count.

One note on signs. Stockfish reports a score from the side to move. Every evaluation below is written from White's point of view: +0.57 means White is ahead by more than half a pawn, and a negative number means Black is better.

Move 2: do players accept the gambit?

After 1.e4 e5 2.f4, Black is choosing between accepting the pawn, declining it, and a handful of quiet developing moves. Stockfish ranks 2...exf4 first at -0.64, then 2...d5 at -0.30, 2...Nc6 at -0.05 and 2...Bc5 at +0.12. So the engine agrees with Fischer's basic instinct: accept the pawn, and Black is the one who is better.

Rating1000120014001600180020002200
Probability of 2...exf4, accepted (best, -0.64)42.7%44.0%43.4%42.9%42.8%42.1%40.1%
Probability of 2...Nc6 (-0.05)23.1%24.1%24.6%23.6%20.7%17.2%14.4%
Probability of 2...d5, the Falkbeer (-0.30)out3.8%5.0%7.5%12.1%19.7%27.8%
Probability of 2...Bc5, the Classical (+0.12)outoutout3.4%4.1%4.8%6.9%

In this table, "out" means the move was outside Maia's top five at that band, not zero probability. The accepting move is the likeliest at every band, and it barely moves: 42.7 percent at 1000, 44.0 at 1200, 40.1 at 2200. About four in ten Black players take the pawn whatever their rating. What changes is the rest. 2...Nc6 shrinks from 24.6 percent at 1400 to 14.4 at 2200, while the Falkbeer climbs from 3.8 percent at 1200 to 27.8 at 2200, and at 2200 it is the second most likely reply. The Classical 2...Bc5 is never above 6.9 percent, and the engine scores it as the only one of these four that leaves White better. For a White player, the likely outcome of 2.f4 is a pawn capture or a strange developing move, not a theoretical decline.

Move 3: after 3.Nf3 the engine move is rarely found

After 2...exf4 3.Nf3, Black has six named third moves on Wikipedia's list: the Fischer Defense 3...d6, the Modern Defense 3...d5, the Cunningham Defense 3...Be7, the Schallopp Defense 3...Nf6, the Becker Defense 3...h6 and the Classical 3...g5. Stockfish puts 3...d6 first at -0.66, then 3...g5 at -0.58, 3...Nf6 at -0.54, 3...h6 at -0.42 and 3...Be7 at -0.32. All of them keep Black ahead.

Rating1000120014001600180020002200
Probability of 3...d6, Fischer (best, -0.66)8.3%11.2%14.5%16.8%16.9%14.7%12.2%
Probability of 3...g5, Classical (-0.58)out8.3%10.5%14.8%22.7%33.7%39.8%
Probability of 3...Nc6 (+0.08)22.9%26.9%25.5%19.6%12.4%outout
Probability of 3...Nf6, Schallopp (-0.54)24.4%18.0%12.8%9.3%out6.6%10.5%
Probability of 3...Be7, Cunningham (-0.32)outoutout9.4%14.1%13.2%7.0%

The engine's choice, Fischer's 3...d6, is found only 8.3 percent of the time at 1000, and tops out at 16.9 percent at 1800. It is never the likeliest move at any band. Instead, the likeliest move is three different moves across the bands. At 1000 it is 3...Nf6 (24.4 percent, -0.54). From 1200 to 1600 it is 3...Nc6, which is the one clear slip in this table: it scores +0.08, so a player who drifts into it has given back the whole Black advantage. From 1800 up it is 3...g5, reaching 39.8 percent at 2200.

That last fact matters, because Fischer called 3...g5 inexact in his article, on the grounds that it gives White drawing chances in the ending. The engine barely separates them: 3...g5 is -0.58 against -0.66 for his own pick, and either is fine for Black. In the same article Fischer said 3...d5 "equalizes easily, but that's all". The piece is often linked to his loss to Boris Spassky in a Kieseritzky Gambit at Mar del Plata in 1960, but that motive is a repeated story we cannot confirm. As of November 2026, Chess.com's database of 381 Fischer Defense games shows Black winning 50 percent and White 35, with no rating filter given.

Move 4: humans play 4.Bc4, the engine plays 4.d4

Black's most popular reply at higher ratings leads here. After 3...g5, White picks a fourth move, and the choice splits the engine from the crowd.

Rating1000120014001600180020002200
Probability of 4.Bc4 (-1.03)32.5%41.0%48.8%55.3%59.3%59.3%54.4%
Probability of 4.d4 (best, -0.68)29.1%25.8%21.5%18.2%15.8%14.1%13.6%
Probability of 4.h4 (-0.93)9.5%12.5%14.9%16.1%17.4%20.6%25.4%

4.Bc4 is the likeliest move at every band and climbs from 32.5 percent at 1000 to 59.3 at 1800 and 2000. The engine's 4.d4, scored at -0.68, falls the other way, from 29.1 percent to 13.6. Stockfish rates 4.Bc4 at -1.03, about a third of a pawn worse than 4.d4. The bishop move is the book move with the best reputation, because it aims at f7 and sets up the Muzio and Salvio ideas below, so players choose it for its attacking story rather than its score. Meanwhile 4.h4 (the Kieseritzky and Allgaier family when Black plays ...g4) grows with rating, to 25.4 percent at 2200. Every fourth move in the engine's top eight is worse than -0.68 for White, down to -1.55 for 4.Be2, so the real conclusion is that White is already paying for the gambit at this point.

Move 5: Muzio, Salvio and the move nobody plays

After 4.Bc4 g4 the position splits two ways. White can castle and sacrifice the knight (5.O-O gxf3 6.Qxf3), the Muzio Gambit, or play 5.Ne5 Qh4+ 6.Kf1, the Salvio Gambit, which Wikipedia (checked November 2026) considers better for Black. Stockfish ranks 5.Nc3 first at -1.11, then 5.Ng1 at -1.21, 5.O-O at -1.25 and 5.Ne5 at -1.35.

Rating1000120014001600180020002200
Probability of 5.Ne5, Salvio (-1.35)59.9%52.6%42.8%35.6%31.1%28.9%27.7%
Probability of 5.O-O, Muzio (-1.25)10.6%18.8%27.4%34.5%41.3%49.4%58.8%
Probability of 5.Bxf7+ (-2.33)6.7%13.3%18.8%21.6%20.9%15.9%8.1%
Probability of 5.Nc3 (best, -1.11)0.3%0.3%0.3%0.3%0.3%0.3%0.4%

The engine's best move, 5.Nc3, is under half a percent at every band. Nobody plays it, and none of the pages we read mention it. The low bands go for the Salvio knight jump, 59.9 percent at 1000, and the share falls as rating rises. The crossover is at 1600, where 5.Ne5 has 35.6 percent and 5.O-O 34.5. From 1800, castling leads, reaching 58.8 percent at 2200. 5.Bxf7+, a bishop sacrifice scored at -2.33, peaks at 21.6 percent at 1600, which means about one in five predicted White moves at that band throws away more than a further pawn against the best move. At 1000, 5.Ng5 takes 7.2 percent and is scored at -3.91.

The numbers are close enough that the Muzio and Salvio are practical choices, not blunders: 5.O-O at -1.25 is a tenth of a pawn better than 5.Ne5 at -1.35, and it matches what stronger players pick. But both are over a pawn down by the engine. White plays them for activity, not for equality.

The Falkbeer: the decline that holds up

If Black answers 2.f4 with 2...d5, White gets the choice again. The older main line of the Falkbeer Countergambit is 3.exd5 e4, with Black sacrificing a pawn for quick development. As of November 2026, Wikipedia says White's 4.d3 gives some advantage there, and that the most common move after 3.exd5 is now 3...c6, the Nimzowitsch-Marshall Countergambit (Marshall beat Teichmann at Ostend in 1905). It also says theory has not reached a definitive verdict. We did not measure either of those lines. Our data follows 3.exd5 Qxd5, where Black recaptures at once with the queen.

Rating1000120014001600180020002200
Move 3: probability of 3.exd5 (best, -0.34)28.4%42.1%52.5%58.4%62.3%65.4%69.9%
Move 3: probability of 3.Nf3 (-0.37)16.1%18.9%21.3%22.3%22.0%20.2%16.2%
Move 3: probability of 3.fxe5 (-4.34)20.8%15.3%10.4%7.9%6.4%4.9%3.4%
After 3...Qxd5: probability of 4.Nc3 (best, +0.57)50.3%59.6%61.5%58.9%57.6%63.9%75.5%
After 3...Qxd5: probability of 4.fxe5 (+0.00)20.8%16.3%15.8%18.6%21.8%20.1%14.2%

Here the engine's move and the human move agree. 3.exd5 is the engine's choice and the likeliest move at every band, from 28.4 percent at 1000 to 69.9 at 2200, and 4.Nc3 is both best and likeliest after the queen recapture, rising from 50.3 percent to 75.5, with a dip at 1800. White's most common mistake is the natural grab 3.fxe5, which would win a pawn on the board but is scored at -4.34, and takes 20.8 percent at 1000. It fades to 3.4 percent at 2200.

Put the two evaluations side by side. With best play after 3.exd5 the position is -0.34, so Black is fine, yet the line with 3...Qxd5 4.Nc3 is +0.57 for White. The queen recapture is therefore not Black's best move, and if you defend with the Falkbeer you should look at the pawn sacrifice lines first. We have no dedicated position for 2...Bc5, only its share in the 2.f4 table above, so we say nothing further about the Classical Defense.

What this means for a club player

As White, the King's Gambit works as a surprise weapon, not as a theoretical equalizer. After 2...exf4, every White line we measured leaves Black ahead by the engine, and your opponent will often not know it: 2...exf4 is the likeliest reply at every band, but 3...d6, the engine's favourite after 3.Nf3, is under 17 percent at all of them. After 3...g5, play 4.d4 (-0.68) if you want the best score. If you want the attacking game, 4.Bc4 g4 5.O-O is a tenth of a pawn better than 5.Ne5 and avoids the Salvio, which Wikipedia calls better for Black. Learn 5.Nc3 too, since it is the engine's choice and nearly nobody knows it. The Bishop's Gambit, 3.Bc4 instead of 3.Nf3, is the line of the Immortal Game, where Anderssen beat Kieseritzky in London in 1851, but we have no data on it, and Wikipedia reports that John Shaw calls 3...Nc6 a refutation of it. If you want other ways to sacrifice early after 1.e4 e5, the Evans Gambit gives a similar attacking game from the Italian, and the Danish Gambit gives up two pawns for development.

As Black, take the pawn. That holds in every gambit in our comparison of chess gambits, and here 2...exf4 is the engine's top move at -0.64, and 3...d6 or 3...g5 keeps the edge, within a tenth of a pawn of each other. Do not drift into 3...Nc6 on move three, which our model says the 1200 to 1600 bands do: it gives back the whole advantage. Against 3.exd5 Qxd5 4.Nc3, expect to be worse, so against the Falkbeer look at the older main line 3...e4, or at 3...c6, instead of recapturing with the queen. Black players who would rather never face 2.f4 can answer 1.e4 with something other than 1...e5; the King's Pawn Opening guide covers the options. For another sharp attack Black can fall into, see what each rating plays in the Fried Liver Attack.

Where other pages are better

Wikipedia has the full line tree, with the Kieseritzky main line 5...Nf6, the Allgaier and every named defence, and it is the better place to learn the structure. Chess.com's openings page has famous games and the Fischer Defense sample. If you want to play the King's Gambit seriously, a dedicated book or video course is a better source than this page, because a repertoire needs the deep lines after 4.d4 and the Falkbeer 3...e4 and 3...c6, which we did not measure.

One warning about forums: the 2012 story that the King's Gambit had been solved as a draw was an April Fools' prank, and it still gets repeated as fact. What none of those pages shows is what each rating band chooses, or an engine score on every legal move. That is the gap this page fills. What it does not cover: real game results, the Bishop's Gambit, the Classical 2...Bc5 beyond its share, and the lines after 3...e4 or 3...c6 in the Falkbeer.

What to take from it

The King's Gambit is a gambit that works because the opponent does not know the answer, not because the engine likes it. Black takes the pawn about four times in ten at every band, the engine's 3...d6 is found by fewer than one in five, humans prefer 4.Bc4 over the stronger 4.d4, and the engine's 5.Nc3 is almost never played. In the Falkbeer, both 3.exd5 and 4.Nc3 are the engine's choice and the share finding them rises with rating. For a club game against someone you expect to be unprepared, it is a fine choice. Against a prepared opponent it costs you a pawn of evaluation. If you want a quieter 1.e4 e5 system to compare it with, see the Italian Game rating by rating.

Maia3 is a model of human choice built from the maia3-5m checkpoint, not a count of real games, and every percentage above should be read that way.

Frequently asked

What is the King's Gambit?
It is the opening 1.e4 e5 2.f4, where White offers the f-pawn to pull Black's e-pawn away from the centre. ECO assigns it ten codes, C30 to C39. Black can accept with 2...exf4, decline with 2...Bc5 (the Classical Defense) or counter with 2...d5 (the Falkbeer Countergambit).
Is the King's Gambit good?
By the engine, not really, but it is playable. In our data Stockfish 18 at depth 20 rates 2...exf4 at -0.64 for White, and the engine's top eight White fourth moves after 3...g5 all score between -0.68 and -1.55. That is a comfortable edge for Black, not a lost game. In a club game the gambit is a practical weapon, because Black has to find several accurate moves in a row.
Did Fischer refute the King's Gambit?
He claimed to. His article "A Bust to the King's Gambit" appeared in the first issue of American Chess Quarterly in 1961, and says "In my opinion the King's Gambit is busted. It loses by force." He recommended 3...d6 against 3.Nf3. No refutation has been proven, though, and Stockfish rates 3...d6 at -0.66 for White, only a little better for Black than the 3...g5 that Fischer called inexact (-0.58).
Was the King's Gambit solved by a computer?
No. A 2 April 2012 ChessBase story said Vasik Rajlich and a Rybka cluster had solved the accepted gambit as a draw. It was an April Fools' prank, and Rajlich later said chess is about 25 orders of magnitude away from solving something like the King's Gambit.
What is the difference between the King's Gambit Accepted and Declined?
Accepted means Black captures with 2...exf4 and holds the pawn for a while. Declined means Black does not capture: 2...Bc5 is the classical way, and 2...d5 is the Falkbeer Countergambit. In our data about 42 to 44 percent of predicted Black moves at 1000 to 2000 are 2...exf4, and the declined moves grow with rating.
Do strong players still play the King's Gambit?
Rarely, but yes. Magnus Carlsen beat Wang Yue with it in 2010, and Ding Liren beat Carlsen in the online Magnus Carlsen Invitational in 2020 after Carlsen opened with it. Hikaru Nakamura has also played it at the top level. It is far more common at club level than among elite players.

Want to try 2.f4 against a player at your rating?

Chessdrive runs the real maia3-5m checkpoint on-device as your opponent at any rating you set from 600 to 2600, so you can play 2.f4 as White, or face it as Black, and find out how a player at your level really responds. Finished games are reviewed separately with Stockfish so you can see exactly where the position turned. Available now on iOS.

Keep reading

All guides

Updated