London System: The Reply Your Rating Actually Plays

On a chess.com thread asking whether the London System is bad for beginners, AngryPuffer does not equivocate: "the london system is quite dry for both colors, so if you really want to enjoy chess, then play something else." A few replies down, tygxc pushes back: "London system is a viable choice, especially if coupled with Caro-Kann Defense and Slav Defense." By the end of the same thread, the original skeptic, Im_a_Crow, has changed their mind entirely: "sorry, I researched a bit. London is the best opening ever."

Nobody in that thread has access to the one thing that would actually settle their argument for their own games: what a real opponent, at their own rating, is likely to play back at them. That is the gap Chessdrive's own Maia3 data fills. We measured two positions from the London System the same way we measured every opening in our chess openings by rating dataset: the maia3-5m checkpoint, see how Maia chess actually works, predicting what a human at each of seven rating bands, 1000 to 2200, is likely to play, graded against what Stockfish actually is, here Stockfish 18 at depth 20 with every legal move scored. The result below is not a repertoire. It is a picture of what your own opponents actually reach for.

Where the name actually comes from

The London System is older than its reputation as a modern beginner's shortcut suggests. Per Wikipedia, the Irish-American master James Mason was the first player to employ it regularly, at the 1882 Vienna tournament, where he finished third, then again at the 1883 London tournament and the 1889 New York tournament. Before 1922 the opening carried his name, the Mason Variation. The name it has now comes from a single strong event, the 1922 London tournament, where it appeared seven times, including games by Jose Raul Capablanca, Alexander Alekhine and Akiba Rubinstein. Wikipedia is specific that the opening was not created to dodge any particular Black defense; it simply showed up often at that one tournament. The same London style setup later became a standard Black reply to the Reti Opening too, notably in the Reti versus Capablanca game at New York 1924.

It then spent most of the twentieth century rare at grandmaster level before becoming a club level staple in the twenty first. Wikipedia names Magnus Carlsen, Gata Kamsky, Levon Aronian, Teimour Radjabov, Vladimir Kramnik and Fabiano Caruana as modern practitioners, and thechessworld.com describes Kamsky specifically as having made it "the primary part of his repertoire for many years in many big-time events," with Boris Grachev named as another player who reaches for it often. Its biggest single moment came in 2023: Ding Liren used it to beat Ian Nepomniachtchi in Game 6 of the World Championship, the first time the opening had appeared in a world title match at all. Nepomniachtchi's 6...Bf5 took Ding out of his own preparation, and he won anyway, with a mating attack that finished 41.d5!! and forced resignation shortly after 44.Qf7 threatened unstoppable mate.

Three ways to reach the same bishop move

The bishop can arrive on f4 more than one way, and which way matters for what White is actually trying to do. Chess.com's own opening pages give two orders against 1...d5: the traditional 1.d4 d5 2.Nf3 Nf6 3.Bf4, and the modern, accelerated 1.d4 d5 2.Bf4 straight away, which chess.com states current theory now slightly favors. Against 1...Nf6 without an early ...d5, the same bishop development falls under a different pair of ECO codes: A46, when Black delays or avoids a kingside fianchetto, and A48, when Black plays an early ...g6 instead, a King's Indian shaped shell aimed straight at White's setup. If facing that exact shell is your actual problem, our own King's Indian Defense findability data covers it from the other side of the board.

A fourth move order swaps the whole plan: the Jobava London, 1.d4 d5 2.Nc3 Nf6 3.Bf4, credited to GM Baadur Jobava, reported to have introduced it in 2013, with Richard Rapport credited alongside him for developing its modern strategic ideas, though that exact year and sole attribution rest on secondary sources rather than a primary tournament record. Committing the knight to c3 before the bishop move unlocks a thematic Nb5 jump at Black's c7 square and a Qd2 to Bh6 plan against a fianchetto, ideas the standard move order does not have available in the same way. Chessigma treats it, along with the related Barry Attack, where White meets a ...g6 fianchetto with an early Nf3 and Nc3 alongside Bf4, as a distinct system from the plain London it covers on its own trainer page. None of that is what our own dataset measures. Both positions below start from the direct 2.Bf4 with no early Nc3 anywhere, so nothing here tells you what a Jobava player's opponents actually find.

Position A: the best reply almost nobody reaches for

After 1.d4 d5 2.Bf4, Black has 27 legal moves. Stockfish's own ranking is not close at the top: 2...c5 is best at +0.12, then Bf5 at +0.20, Nf6 and e6 tied at +0.22, and g6 at +0.25, all within 0.13 of a pawn of one another. Further down the list sit Nc6 at +0.42, c6 at +0.48 and f6 at +0.70, each measurably worse. Every single one of those evaluations favors White, which is the first honest thing to say about the London: nothing Black plays here actually equalizes by the engine's own count. The real question is not whether White is better. It is by how little, and whether a human at a given rating can find the move that keeps that margin smallest.

Read the top row first. 2...c5, the objectively best move on the board, rises at every single rating step we measured, with no exceptions: 2.6 percent at 1000, past 9 percent at 2000, up to 13.6 percent by 2200. That is a genuinely clean climb. It is also, at every one of those seven bands, never the single most common reply in the room. Chessigma's own dedicated Anti-London page already recommends exactly this move, calling it "the standard club recipe" and stating that "below 1600 it turns a quiet, comfortable White opening into a fight." ChessAtlas's 2026 guide to beating the London agrees, naming the c5, Nc6, Qb6 line as the setup a player under 1500 should learn first. Both are right about which move is best. Neither says how rarely a real opponent at any of these ratings actually reaches for it, which is the specific thing our own data adds.

What the field actually plays instead has its own shape. 2...Nc6 opens as the single most common reply anywhere on the table, 40.0 percent at 1000, and stays the leader through 1400. Then it fades hard, down to 13.9 percent by 1800, and drops out of the top five entirely by 2000 and 2200. 2...Nf6 runs the opposite direction, starting as only the second choice at 20.0 percent, climbing past Nc6 by 1600, and reaching 38.7 percent at 2200, the single most common reply at the top of our range. e6 holds steady behind the leaders at every band, between 13.4 and 20.0 percent. The late arrival is 2...c6, which does not make the top five until 1800 and then outdraws c5 itself at 2200, 15.7 percent to 13.6, despite sitting more than a third of a pawn worse by engine count. The practical read for a London player is not complicated: expect Nc6 from a weaker opponent and Nf6 from a stronger one, and expect c5, the move every serious guide recommends, from almost nobody at any rating you are likely to actually face.

How much does missing c5 actually cost Black? Less than the gap in popularity suggests. Stockfish's top five, c5, Bf5, Nf6, e6 and g6, all sit within 0.13 of a pawn of one another, so a player choosing between them is not stumbling into a bad position. The exception is the move the lower bands like best. 2...Nc6 gives up three tenths of a pawn against c5, and it is the single most likely reply at 1000, 1200 and 1400. From 1600 up, the most likely reply is Nf6, and the cost drops to a tenth. That is the number behind a claim London players make defensively in forum threads: against a strong opponent, the opening's reputation for being low risk for White is not just talk. Against a 1200, the most popular answer hands White a little extra for free.

Position B: the reply that gets harder to find as players get stronger

Play through the most common human move at each rating in Position A and you mostly reach 2...Nc6, so that is the line worth following one ply deeper: 3.e3, the natural continuation and, per our data, the most likely human move at 1400. After 2...Nc6 3.e3, Black has 28 legal moves. Stockfish's ranking here still keeps White ahead everywhere: Nf6 is best at +0.43, then e6 at +0.48, Bf5 at +0.53, a6 at +0.68, f6 at +0.72, and, well clear of the rest, e5 at +1.05.

This table runs backward from Position A's, and it is worth sitting with why. Nf6 is Stockfish's own top choice, and it is also the single most common human reply from 1000 through 1400. Rather than growing more findable with rating the way c5 did one ply earlier, it shrinks: 39.4 percent at 1000 down to 30.7 percent at 2000, essentially flat from there at 31.7 percent by 2200. Bf5, a mere tenth of a pawn worse by engine count, overtakes it at 1600 and keeps climbing to 39.3 percent, second only to Nc6's 40.0 at 1000 across both tables. A 2200 rated opponent is measurably less likely to play the engine's move here than a 1400 rated one is. That is not a data error. Stronger players are simply converging on a different, nearly equally good move.

3...e5 is the one clearly bad choice on this table, costing more than half a pawn against Nf6's +0.43. It shows up at 4.1 percent at 1000 and is gone from the top five everywhere above that, the clearest sign in either position of a beginner habit that rating actually cures. 3...f6, a slower plan aimed at an eventual ...e5 break, moves in neither direction cleanly: 12.3 percent at 1000, down to 6.8 percent at 1600, then back up to 12.5 percent at 2200, a genuine U shape our data does not have a tidy explanation for.

Here is the honest complication with this whole position. It only exists because White reached 3.e3 after 2...Nc6, and 2...Nc6 is the move a 1400 rated Black player reaches for, not a 2000 or 2200 rated one. By 2000, Nc6 is not even in Position A's own top five anymore. Position B, in other words, describes a game your stronger opponents mostly will not give you. Treat it as what club level games below about 1800 actually produce, not as the position to expect against the strongest players you will actually meet.

One flat number was hiding two opposite trends

Chessdrive's own chess AI findability rankings already report a single blended figure for the London System: 21.0 percent at 1000, rising to 22.6 percent at 2200, a line the page itself describes as barely moving at all. That number is not wrong. It is the average of the probability that a human finds Stockfish's best move in each of the two positions above: 2.6 plus 39.4, divided by two, at 1000; 13.6 plus 31.7, divided by two, at 2200. Averaging together a line that climbs and a line that falls produces almost no visible movement, which is exactly what that single blended number shows. It also hides the entire point of this article: one position gets easier for Black to answer correctly as rating rises, and the very next one gets harder, at the same time, and a reader who only sees the blended 21 to 23 percent figure would never know either was happening.

What to actually play against each common reply

Against 2...c5, the reply every competitor guide correctly recommends and almost nobody under 2000 actually plays, Chessigma's own Anti-London page lays out the idea plainly: c5 "jabs the base of their center" and hits the b2 square left soft once the bishop has left c1, usually continuing with ...Nc6 and ...Qb6. What none of the guides that recommend it quite say is that the sharpest branch underneath it is not settled even among strong club analysts, as the pfren versus tygxc disagreement over 4...Qxb2 above shows directly. Whichever side of that argument you land on, "the recommended answer" and "a line that is simple to play" are not the same claim, and nobody has fully reconciled them.

Against 2...Nc6 and 3...Nf6, the two moves your opponent is most likely to actually play below about 1800, there is no trap to spring, only ordinary development: e3, Nf3, Bd3 or Be2 and O-O, with c3 to shore up d4 and a knight heading for e5 once your pieces are out. One warning is worth taking from chess-teacher.com's own London guide: once Black has fianchettoed with an early ...g6, an automatic Bd3 is a mistake, because "...e5!" then wins material outright. Check where your bishop actually wants to sit before moving it on autopilot.

Against 3...Bf5, the move overtaking Nf6 in our own data from 1600 up, Black has developed the light squared bishop outside the pawn chain before playing ...e6. The two standard answers are Bd3, offering to trade that bishop off, and Qb3, hitting the b7 pawn the bishop has just stopped defending. Against 2...e6 or an early ...g6 fianchetto, expect a slower game with a King's Indian or Queen's Indian flavor rather than a sharp one; if the fianchetto itself, not the London specifically, is the recurring problem in your games, our King's Indian Defense article covers what that setup looks like from the other side of the board. None of this is a memorized line, and it should not be treated as one. If you would rather understand why a plan works than memorize the moves that produce it, that is the entire argument of stop memorizing openings and learn the plan instead.

Where the London genuinely comes up short

None of the above should read as an endorsement without limits. The London's reputation problem is real and comes from players who have actually faced it, not just from marketing copy on the other side. On a chess.com thread asking why the opening gets so much hate, SamuelAjedrez95 calls it "an opening designed so that you can practically play it with your eyes closed and ignore the opponent," and BOWTOTHETOAST adds that "it awards passivity and timidity." On a separate thread, NikkiLikeChikki is blunter still: "It's a crime against chess." Frederic91, on a third thread, makes a more specific complaint, that it "pratically skips the opening step to go right to the middlegame and that's why is not considered particularly instructive," and Kowarenai calls it "common but kind of drawish or messy if you dont know how to use it."

Even a self described seven year London player will not claim it is objectively strong. Asked directly whether 1.d4 followed by Bf4 gives White a real edge, GM Kirk Ghazarian answers "in short, my honest answer is .... no!", framing his broader view as "black is equal with correct play against every opening," not a London specific claim at all. A poster on a lichess.org thread reaches almost the identical conclusion from the other side of the board: "The London is meh. However, it is meh only because black can get close to equality out of the opening."

The defense of the opening is just as real, and it is not only "it is easy." Laskersnephew points to who actually plays it: "Almost every great player has played the London System as white. Carlsen, Nakamura, Gata Kamsky." ChessDoofus, a National Master, reports a practical result rather than a reputation: "I've beaten, let's see here, five masters with the London in rated tournaments." Chess.com's own master database, read in September 2026, cuts the other way. Against the early c5 counterattack, chess.com reports that "Black achieves 49% of the games, drawing 23%, and losing only 22%," and against the Indian Setup with an early ...g6, Black scores 38 percent wins, 31 percent draws and 31 percent losses across more than 2,000 games. Those are results from real master games, a different measurement from the move probabilities above, and they are the strongest argument on this page for learning 2...c5, the reply so few players at any rating actually reach for.

One more honest note on scale. Firechess's analysis of roughly 1.5 million Lichess games, as read in September 2026, found the London getting less common as rating rises, from 4.8 percent of games under 1400 down to 3.5 percent at 1800 and above. That measures something different from everything else in this article: how often White chooses to play the London at all, not how Black replies once White has. Do not read the two together as one trend. A club player's opponents are simply less likely to open with 2.Bf4 in the first place as the room gets stronger, and our own data only describes what happens in the games where they still do.

What this data is, and what it is not

Maia3 predicts what a human at a given rating is likely to play. It is not a record of real games, and its own project description says as much, describing it as reproducing "common human biases" and making "many of the same mistakes that humans make." Every Maia3 percentage in this article is a model of human tendency, not a box score.

The scope is narrower than "the London System" as a whole, too. This dataset measures exactly two positions: the tabiya after 2.Bf4, and the one three ply later after 2...Nc6 3.e3. It says nothing about the Jobava London, the Barry Attack, the ...g6 fianchetto setups covered in our King's Indian Defense data, or any branch of the c5 counterattack past Black's own second move. Position B's shape is not unique to the London, either. Our own Caro-Kann Defense findability data found the same kind of crossover on Black's fourth move, one named main line overtaking another at a single rating, while in the King's Indian the engine's move stays the crowd favourite and simply loses share. The same bishop move also meets 1...f5, and our Dutch Defense page shows 2.Bf4 is the likeliest White move at 1000 and 1200.

To see every position in this article laid out on an actual board, move by move, at your own rating band, Chessdrive's interactive London System page renders exactly that. And if you want to check whether you found the engine's move in your own last London game rather than guess from a table, Chessdrive's chess analysis board will tell you directly.