Benko Gambit: Why Club Players Decline It With 4.b3

The Benko Gambit is 1.d4 Nf6 2.c4 c5 3.d5 b5, and it is one of the few gambits in which the pawn is given for a plan rather than an attack. Guides explain the plan and list the White tries. They do not tell you which White reply you will actually meet at your level. We ran the maia3-5m checkpoint, see how Maia chess actually works, on two positions in the Benko at seven rating bands from 1000 to 2200, and graded every White move with Stockfish 18. The result is a clean surprise: at club level White usually does not take the pawn. Percentages on this page are Maia3's predicted probability that a player at that rating chooses the move. They are not game counts.

One convention before the numbers. Every evaluation is from White's side, so +0.66 means White is better by 0.66 of a pawn. Our tables use Stockfish at depth 20. The named line checks use a separate local run at depth 22, so the same position can differ by a few hundredths between the two.

What the Benko Gambit is

After 3.d5 Black plays 3...b5, offering the b pawn. Wikipedia puts the opening under ECO codes A57 to A59, with A59 the main accepted line. The USCF Sales blog describes it as more positional in nature than most gambits, which trade material for a quick attack. Black is paying a pawn for open lines on the queenside and a bishop on the long diagonal, and what Black gets does not fade as pieces come off. Wikipedia notes that Black does not generally mind if queens are exchanged, and one forum player says the same, that Black seems to have compensation even without queens.

Where it came from, and who played it

The name is not universal. In Eastern Europe it is the Volga Gambit, after an article in a 1946 Russian chess magazine. Wikipedia credits Karel Opocensky with earlier versions of the idea in the 1930s, and says the first game with the now standard move order was Thorvaldsson against Vaitonis at the Munich Olympiad in 1936. English speaking countries use the name of Pal Benko, who popularised it from the 1960s and published a book on it in 1974.

Benko was born in Amiens, France, in 1928 and died in Budapest in 2019, aged 91. He was Hungarian until 1962, then American, took the grandmaster title in 1958 and won or tied first in a record eight U.S. Open Championships. He also gave up his 1970 Interzonal place to Bobby Fischer. ChessBase quotes his autobiography saying the gambit was first regarded as a joke by his peers, and that practice has shown it to be a dangerous weapon which is easy to learn.

Wikipedia lists former world champions Anand, Kasparov, Topalov and Carlsen among players of the line, plus Ivanchuk and Gelfand. Forum opinion is less sure about its place at the very top. One poster calls it more or less sound, another says the Marshall looks a lot sounder these days, and a third says Kasparov played it but not in high level games. We did not settle that argument and do not claim to. Wikipedia says it remains popular below master level.

The key moment: how White answers 3...b5

After 3...b5, Stockfish 18 at depth 20 ranks White's options like this: 4.cxb5 at +0.66, 4.Nf3 at +0.35, 4.Qc2 at +0.33, 4.e3 at +0.22 and 4.Nc3 at -0.29. Taking the pawn is clearly the engine's choice. 4.b3 scores -0.18 in the same run, 84 centipawns lower. Our depth 22 check of the same position gives 4.cxb5 at +0.73; the extra depth accounts for the difference.

Here is what Maia3 predicts people play.

In the table, "out" means the move was outside Maia's top five at that band. It does not mean zero probability. Read the first two rows against each other. From 1000 to 1600 the likeliest White reply is 4.b3, at 33.2, 37.3, 40.1 and 40.1 percent, and the engine's 4.cxb5 sits second. At 1800 the two are a near tie, 34.9 percent for 4.cxb5 against 34.4 for 4.b3. Then taking the pawn pulls away, to 53.0 percent at 2000 and 70.4 at 2200, where 4.b3 has fallen to 4.7.

The gap is not a footnote. Across the 13 openings in our second dataset, the Benko has the largest mean evaluation gap between the engine's best move and the likeliest human move at 1000 to 1600. That mean is 42 centipawns, averaged over two positions, the early one after 3...b5 and a later one after 4.b3 bxc4, where the gap is zero. The 84 centipawn figure belongs to the early position alone. Our chess openings by rating dataset uses the same method for other openings.

Why would club players decline? The data does not say, and we will not pretend to know. A reasonable guess is that giving back a pawn is what most players first learn to avoid, and 4.b3 keeps the c4 pawn from being taken and the position familiar. Treat that as a guess only. What the numbers show is the behaviour, and the behaviour is the point of this page: if you play the Benko below 1800, you will meet 4.b3 more often than the textbook 4.cxb5.

What 4.b3 actually concedes

The pawn gets traded off, not defended. After 4.b3 bxc4, Maia3 predicts 5.bxc4 at 99.5 percent at 1000, 99.6 at 1200 and 99.7 at 1400 and every band above. The dataset scores the position after 5.bxc4 at +0.08 for White, with the next best White move, 5.Nc3, much lower at -0.43. Since almost everyone recaptures, there is little to prepare in the reply.

Our depth 22 check of 4.b3 itself gives +0.01, with 4...bxc4 as Black's best reply and 4...e6 close behind at +0.04. After 4...bxc4 5.bxc4 it scores +0.03, and the engine's line runs 5...g6 6.Nc3 Bg7 7.Rb1 O-O 8.e4. This is a small difference from the dataset's -0.18 for 4.b3, because the two runs search to different depths. Either way, the verdict is the same. White has given up the edge that 4.cxb5 keeps, and Black is fine.

For Black the plan is simple. Capture, recapture, play ...g6 and ...Bg7, put a rook on the b file and a knight on a good square, and aim for the pressure you wanted all along. You are not worse, and there is no need to hunt for a gambit that no longer exists. The other declines also give up part of White's edge. Our depth 22 check scores 4.a4 at +0.07, with 4...Qa5 as Black's reply, and 4.Nf3 at +0.45, with 4...b4 answering, to block the c3 square. The USCF Sales blog calls 4.Nf3 rare and says it gives Black good play, and our numbers agree it is more modest for White than taking the pawn at +0.73.

The accepted lines, with engine numbers

When White takes, the usual run is 4.cxb5 a6. Our depth 22 check puts the position after 4...a6 at +0.68, with 5.e3 as the best move. Every score from here is from our local run at depth 22. The main branches are the following.

5.bxa6 Bxa6 is the main line, and Wikipedia describes the result as fast development, control of the long diagonal and pressure down the open a and b files. The position scores +0.87, with 6.Nc3 White's best. After 6.Nc3 d6 7.e4 Bxf1 8.Kxf1 g6, Black has traded off White's light squared bishop at the cost of White's castling rights. The engine gives +0.88, with 9.g3 first and 9.a4 close behind at +0.85. White reorganises with an artificial castle, Kg2 and g3, and keeps a small, steady edge. The USCF Sales blog gives a delayed recapture, 5.bxa6 g6 6.Nc3 Bxa6 7.e4 Bxf1 8.Kxf1, as another route to the same structure. Our check of 5.bxa6 g6 puts the position at +0.77, so the order does not change much.

The Zaitsev Variation is 5.Nc3. We use the name only as a label for the line. It scores +0.49, with 5...axb5 as Black's best reply, which is lower than the main line.

5.e3 scores +0.66 with 5...Bb7 first. 5.f3 is the quietest try at +0.14, with 5...axb5 best. 5.b6 pushes the pawn on instead of capturing, and it scores +0.16, with 5...e6 as Black's reply. Both are far below the +0.87 of the main line. Chess.com says that 5.bxa6 limits Black's queenside pressure, and the engine agrees that the main line is among White's strongest tries, at +0.87 after 5...Bxa6.

Chess.com also lists the fully accepted branches 7.e4, 7.g3 and 7.Nf3, and calls 7.g3 White's best move statistically. It states no pool or rating band, so we leave that as a claim we cannot check. The line we did test was the 7.e4 branch above.

How good is it by the engine

The honest summary is that White keeps a clear but modest edge in the main line, and a smaller one when White declines. At depth 22, taking the pawn gives +0.73 at move four and +0.87 after the recapture on a6. The accepted main line reaches +0.88 after 8...g6. In pawn terms that is a bit under a pawn. The engine does not see full compensation for the pawn. A human player sees more of the plan, the open files and the long diagonal, than a score can show.

That is the drawback with any positional sacrifice. Chess.com calls the gambit sound with straightforward development, and also names its drawbacks, which are that White can decline several ways and sometimes regains the initiative by returning material. As of October 2026, Chess.com also notes that 3...b5 is Black's most popular choice after 3.d5, with 2...c5 Black's third most popular move in its position. It states no pool or rating band, so we cannot say how that varies by level. Our depth 22 numbers say the same thing in another form. The Benko is a defence that asks White to know what to do, and the data show that at club level White often chooses the quieter 4.b3.

What to play by rating band: Black

Below 1800, expect 4.b3 more often than 4.cxb5, 40.1 against 24.3 percent at 1400, and both far more often than 4.e3. Have an answer ready for each, and treat 4...bxc4 as your default against 4.b3. Learn the main line 4.cxb5 a6 5.bxa6 Bxa6 6.Nc3 d6 7.e4 Bxf1 8.Kxf1 g6 well enough to know why each move is played. The idea is clear: trade the light squared bishop, give up the pawn, and take the open files. A forum player points out that Bxa6 stops White from developing the king's bishop in the normal way, and another says the d3 square often becomes White's weak point as Black's pieces come in behind the pawn front. Those are opinions, but they describe the plan well.

At 1800 and above, 4.cxb5 is the likeliest move and its share grows to 70.4 percent at 2200. So prepare the main line first and the 4.b3 reply second. Expect better knowledge from White, including 5.e3 and 5.Nc3, and know which of your replies fits. If you prefer a quieter style, look at the King's Indian Defense page, which shares the fianchetto bishop but asks for no pawn sacrifice.

If you like gambits for their own sake, the Englund Gambit page covers a sharper sacrifice against 1.d4. For learning habits instead of lines, why you should stop memorizing openings and learn plans fits the Benko well, since its plan is the same in nearly every branch. For a different way to unbalance 1.d4, the Dutch Defense page shows what White plays against 1...f5 at each rating.

What to play by rating band: White

If you are White and see 3...b5, the engine says take with 4.cxb5. At depth 22 the best way to keep an edge is 4...a6 5.e3 or 5.bxa6 Bxa6 6.Nc3, and the main line after 8...g6 offers 9.g3 as a safe plan. Below 1800 your opponents will also meet 4.cxb5 less often than 4.b3 by the Maia3 numbers, so a Benko player may be less ready for it than for the move they expect. For the reverse situation, a 1.d4 opening where White is the one offering the pawn, see the Queen's Gambit page.

If you decline, know what it costs. In our depth 22 checks 4.b3 and 4.a4 are close to equal, at +0.01 and +0.07, and 4.Nf3 keeps more at +0.45, still well short of the +0.73 that taking keeps.

Where other pages are better

This page is narrow on purpose. The dataset holds two positions, the one after 3...b5 and the one after 4.b3 bxc4, and nothing past move five. We do not measure how often players reach the accepted main line, we have no game results by rating, and the Lichess explorer, which would give rating filtered results, was not available to us. Wikipedia is the better reference for history and the full list of variations, and chess.com is better for browsing the lines. A forum poster mentions an Epishin variation that makes the gambit a frustrating ordeal, but none of our sources define it, so we do not describe it. None of the pages we opened splits White's replies to 3...b5 by rating, which is the gap this page fills.

What to take from it

The Benko Gambit is a pawn sacrifice for a plan, and the engine says White keeps about +0.9 in the accepted main line. The headline is human, not theoretical: from 1000 to 1600 the likeliest reply to 3...b5 is 4.b3, at 33.2 to 40.1 percent, an 84 centipawn gap to the engine's 4.cxb5, and the pawn is simply given back after 4...bxc4 5.bxc4. Learn the main line for the players who take, and know that below 1800 you will meet the decline first. 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.