Sicilian Defense: The Coin Flip That Never Resolves

"No it is not, End off! It only suits players with a top class(or atleast intermediate) level." That is one reply on a chess.com thread asking whether the Sicilian Defense is playable for a beginner. On the very same site, a different player writes: "I'm a beginner who started last January. I've played the Sicilian defense against 1. e4 every time I got the chance. I went from 800 points to >1400 using it." Both people are describing the same opening. Neither one is wrong.

We built our chess openings by rating dataset to explain exactly this kind of disagreement: how often a real human at a given rating actually finds the engine's best move, position by position, using Maia3 to model human tendency and Stockfish to grade it. Applied to the Sicilian specifically, that data splits into two very different chapters. One position gets more findable at every single rating band we tested, with no exceptions anywhere. One ply later, the fork that decides which named Sicilian you end up playing is a genuine coin flip at 1000, and it still has not fully resolved by 2200.

The move a lot of Sicilian players never actually get to make

Before any of that data matters, there is a more basic problem worth stating plainly before the numbers arrive. The Sicilian this article measures is the Open Sicilian, reached after 2.Nf3 and 3.d4. A large share of players who open with 1...c5 never see it, because White has several ways to sidestep the fight entirely: the Closed Sicilian (2.Nc3), the Alapin (2.c3), the Grand Prix Attack (2.f4), the Smith-Morra Gambit (2.d4 cxd4 3.c3), the Rossolimo (2.Nf3 Nc6 3.Bb5), and the plain Bowdler Attack (2.Bc4). On a chess.com thread, poster AdorableMogwai puts it bluntly: "By far the most popular white response in the <1500 range is 2.Bc4, or the 'Bowdler attack'. The other anti-Sicilians are also popular, and your likelihood of playing against one of them is greater than getting an open Sicilian."

How often does that actually happen? Here the sources genuinely disagree, and we are printing all four rather than picking one. House of Staunton states the Sicilian overall "is only played in 18% of club-level games versus 1.e4," even though it remains, in the same breath, "the most popular Black opening at master level." That is a stat about how often Black chooses 1...c5 at all, not about how often the game then continues into the Open line specifically. USCF Sales, citing Lichess data restricted to amateur ratings 1000 to 2200, reports the Open Sicilian scoring 48 percent for White and 48 percent for Black, against 50/45 for the Grand Prix Attack, 50/46 for the Alapin and 51/45 for the Smith-Morra, and uses that gap to recommend club players adopt an Anti-Sicilian rather than memorize Open theory. That is a scoring stat conditional on already reaching the line, a different measurement again. A third site, MyChessPosters, citing a stated four million plus game Lichess dataset, reports by one account that Anti-Sicilians generally score better for White at amateur level, and that at 1600 to 2000 Elo both the Najdorf and the Closed Sicilian show 48 percent for Black against 46 percent for White, though we could not fetch that page directly to confirm it ourselves. And Lichess's own opening explorer, fetched directly for this piece, shows roughly 54 percent of games after 1...c5 continuing with 2.Nf3, under whatever rating filter the page defaulted to at the time. These four numbers do not answer the same question, and we are not averaging them into one. What every one of them agrees on, from a different angle each time, is that a meaningful share of Sicilian games at club level never reach the position the rest of this article is about.

Real players describe the same gap from inside their own games. On a different chess.com thread, FrogCDE writes: "Most club players up to around 1800 avoid the open Sicilian because they're afraid of the complications." Another poster on that thread, benl123, warns a reader who books up on named lines that there are "a lot of variations to learn in addition to your Dragon, Najdorf etc, which you will hardly get to play." On a thread literally asking whether everyone hates the Anti-Sicilians, ThrillerFan reports the Open lines were so rare for him that he quit playing the opening over it, writing "The only problem was 2.Nf3 and 3.d4, so I quit playing the Sicilian!" and SwimmerBill adds: "So many play anti's these days and so few play main lines that I'm getting better at anti's than at main lines." Keep all of this in mind for everything that follows: the data below describes one specific, heavily analyzed tabiya, not the average Sicilian game a club player actually experiences.

The tabiya we measured, and the cleanest result in the dataset

The line is 1.e4 c5 2.Nf3 d6 3.d4 cxd4 4.Nxd4, the standard route into the Open Sicilian. Wikipedia's own ECO classification runs the whole Sicilian family from B20 through B99, and by one account, after 2.Nf3, Black's most common replies are 2...d6, 2...Nc6 and 2...e6, with 2...d6 the single most common, the move order that leads exactly here. GM Jonathan Rowson frames the underlying trade Black makes with 1...c5 this way: "I'll let you open the position and develop aggressively, but you must give me one of your center pawns." The opening was not always considered sound. Jose Raul Capablanca once called it a defense where "Black's game is full of holes," and Siegbert Tarrasch dismissed it as "not strictly correct." Both verdicts were overturned by decades of practice, a revival that ran through Mikhail Botvinnik and Miguel Najdorf in the 1940s and 1950s, and by 1990 the Sicilian had become, per Wikipedia, "the most played and most analysed opening at both the club and master levels." At master level today, roughly 80 percent of games beginning 1.e4 c5 continue 2.Nf3, and Wikipedia describes Magnus Carlsen as "a regular practitioner of the Sicilian" at the very top of the game.

We measured this position the same way we measured every opening in our chess openings by rating dataset: how Maia chess actually works, specifically the maia3-5m checkpoint, queried at seven self-elo bands from 1000 to 2200 against what Stockfish actually is, here Stockfish 18 at depth 20, with MultiPV set to the exact legal move count of the position so every legal reply gets its own score, not just a top handful.

After 4.Nxd4, Black has 28 legal moves. Stockfish's best is 4...Nf6, at -0.39. Maia3's probability that a human actually plays it rises at every single rating step we tested, with no exceptions:

47.0 percent at 1000 to 93.8 percent at 2200: this is about as clean a result as anything in our full openings dataset produces. The eval gap, the difference between what the engine wants and what the most likely human move actually is, sits at exactly zero at every one of the seven bands, meaning the crowd's favorite move already matches the engine's favorite move even at 1000. The runner-up, 4...Nc6, is worse by Stockfish's count at -0.54, and it runs the opposite direction, falling from 27.0 percent at 1000 to just 1.9 percent by 2200. The shape could not be simpler: the higher the rating, the more certain the field gets that Nf6 is correct, and nobody's peers are drifting the wrong way at any point along the curve.

One ply later, a coin flip that never fully resolves

Position A is the easy part. One ply after 4...Nf6, White replies 5.Nc3, reaching the fork every serious Sicilian player eventually has to answer: Najdorf or Dragon. This is the exact move order Wikipedia gives for both named systems, 5...a6 for the Najdorf and 5...g6 for the Dragon, alongside 5...Nc6 for the Classical and 5...e6 for the Scheveningen. Black has 31 legal moves here, and Stockfish's top choice is 5...a6, the Najdorf move, at -0.33.

At 1000, this is not a findable "best move" the way Nf6 was one ply earlier. It is a coin flip: a6 at 30.2 percent against the Dragon's g6, thirty centipawns worse by Stockfish's count at -0.63, sitting at 30.1 percent, a gap of one tenth of one point. Human tendency does not distinguish between a move that is objectively best and one that gives up thirty centipawns until well past the beginner range.

a6's popularity does climb from there, and climbs a long way: 50.3 percent at 1200, 59.7 percent at 1400, 63.9 percent at 1600, 66.7 percent at 1800, peaking at 67.0 percent at 2000. Then it falls, to 62.8 percent at 2200, its lowest reading since the 1400 band. This is not the clean, one-directional climb Position A produced. a6 stays the single most popular move at every band we tested, so the eval gap still reads zero the whole way through: Stockfish's favorite and the crowd's favorite never actually part ways. But the crowd's confidence in it stops growing after 2000 and gives a little back, and the Dragon try never goes away, sitting at 23.3 percent even at 2200, still taking roughly a quarter of the probability mass at the strongest band we measured. A 2200 rated Sicilian player is more certain about the Najdorf than a 1000 is, but nowhere near as certain as Position A's Nf6 curve would suggest, and a real minority of even strong club players are still reaching for the Dragon instead.

That thirty centipawn gap is real, but it is not a verdict on either opening as a whole. It is one engine's read of one position, one ply deep into two entire named systems each carrying decades of theory. What it does say plainly is that the reputational gap between "the Najdorf is the main line" and "the Dragon is second choice" corresponds to a genuinely narrow, and at 1000 essentially nonexistent, difference in what human players actually reach for over the board. This is not the only reversal our full dataset produces either; our Caro-Kann Defense findability data shows a comparable pattern one opening over, though there the crowd's favorite move actually flips outright rather than just losing momentum.

The two lines earn their reputations honestly. Modern Chess Openings calls the Najdorf the "Cadillac" or "Rolls-Royce" of chess openings, and its list of heavy practitioners includes Bobby Fischer, Garry Kasparov and Maxime Vachier-Lagrave. That reputation comes with a cost even Miguel Najdorf himself complained about: "Here comes some kid who's memorized the moves and he kills me." The Dragon has its own extreme: Fischer's own assessment of the Yugoslav Attack White plays against it was that "White's attack almost plays itself... weak players even beat Grandmasters with it," and Wikipedia calls the Yugoslav Attack itself "one of the sharpest of all chess openings." On a chess.com thread comparing the two directly, IM pfren says they are difficult for different reasons, "The Najdorf due to its positional complexity, and the Dragon because of the overwhelming amount of theory," though pfren also adds "The Dragon is much easier to understand for a post-beginner." Another poster on a related thread, FizzyBand, pushes back specifically for lower ratings: "The tactics and theory of the Dragon is not so good if you are under 1600 or so." Asked directly whether the Dragon is refuted, Robert_New_Alekhine draws a rating line of his own: "At the club-level it is not refuted, but on the top level, I would say yes. It is too risky." Nobody in any of these threads agrees on which named line is actually easier, and our data does not settle that argument either. It only tells you how often your peers are choosing each one.

What to actually do at your own rating

Under about 1400, the fork barely matters yet, because the coin flip is still close to even. Play whichever of a6 or g6 you find more comfortable to learn; the objective thirty centipawn gap between them is small next to everything else that decides a club game. What matters more at this stage is simply reaching the position at all, which, per the section above, is not guaranteed: expect the Bowdler Attack, the Alapin or something else entirely more often than the exact tabiya this article measures.

Between 1400 and 2000, a6 is pulling steadily away from g6 among your peers, climbing from 59.7 percent to 67.0 percent over that stretch. This is the window where committing to the Najdorf starts to pay off in practical terms: more of the people you are actually playing against will also know it, so preparation time stops going toward lines you will rarely see. If you would rather absorb the reasoning behind a6 than memorize the moves that follow it, that is the exact approach we lay out in stop memorizing openings and learn the plan instead.

Past 2000, know that the field's certainty has already peaked and eased back slightly, and that a real minority, close to a quarter of players even at 2200, still play the Dragon instead. If you are one of them, House of Staunton's own advice is worth hearing even though it is a blanket threshold rather than a rating specific to you: it recommends against picking up heavily theoretical lines like the Najdorf before roughly 2000 Elo in the first place, which, if you had followed it, would put you right at the peak of this very curve. And if you want to check whether you actually found the engine's move in your own last Sicilian game rather than guess from a probability table, Chessdrive's chess analysis board will tell you directly.

What this data is, and what it deliberately is not

Maia3 is a model trained to predict what a human at a chosen strength is likely to play, not a tally of real games. Its own project description says it reproduces "common human biases and makes many of the same mistakes that humans make," and that is the entire premise behind every percentage in this article: a validated prediction of human tendency, not a box score of games actually played. We did not query a real game database for either position in this piece of work, and we are not going to imply otherwise.

The other limit is narrower and just as important: this data measures exactly one Open Sicilian tabiya, the line running through 4.Nxd4 and 5.Nc3. It says nothing about the Anti-Sicilian sidelines covered above, nothing about the Rossolimo or any other 2...Nc6 line that skips ...d6 entirely, and nothing about the e6-Sicilian family, the Taimanov and the Kan, both of which delay the same decision by a different route. Even inside the branches our Position B fork reaches, we only measured the first move of each: the Dragon's own deep theory, Wikipedia's ECO range B70 through B79, and the Najdorf's, B90 through B99, both expand enormously beyond the single ply we scored here. Lichess's own opening explorer already does something we deliberately do not: it filters real aggregate popularity by rating band from 1000 to 2500. What it does not do, and what this dataset exists to add, is judge whether the popular move at a given rating is also the objectively correct one, or how findable that correct move actually is for a specific rating.

Where the Sicilian genuinely loses the argument

None of the above should read as a blanket endorsement, and plenty of the people who actually play this opening do not treat it as one. On the same beginner-focused chess.com thread that opened this article, other replies are blunter than the data above: "The Sicilian is way too difficult for a beginner. Should play something like 1. e4 e5," reads one. "No it is not, End off! It only suits players with a top class(or atleast intermediate) level," reads another. Compadre_J, on that same thread, states a specific line: "I don't recommend the Sicilian Defense to Beginners (Players below 1,400 rating)." House of Staunton's own guidance, cited above for the Anti-Sicilian gap, doubles as a theory-burden warning: stay away from the Najdorf specifically before roughly 2000 Elo.

The theory burden is the strongest version of this objection, and it is not exaggerated. Even Miguel Najdorf, the player the whole system is named after, said as much about facing it late in life. If a fully booked opponent reaches the Najdorf or the Dragon against you and you have not, per benl123 above, actually studied "your Dragon, Najdorf etc," the complications this opening is famous for cut against you rather than for you. There is also a structural cost some players never enjoy paying: on one of the same threads, a different reply notes plainly that playing the black side means being "comfortable not castling, which can be tough," since several main lines leave the king in the center for a long stretch by design.

If your actual goal is a lower maintenance opening rather than maximum complexity, the same threads that defend the Sicilian also point away from its deepest lines specifically. On a thread asking what a beginner should actually study inside the Sicilian family, lostpawn247 recommends the Kan or the Accelerated and Hyper-Accelerated Dragon instead of the full Open tabiya, and Jake905 makes a similar case for an e6-Sicilian, calling it "beginner friendly for being practical and flexible considering many club players go with an anti-Sicilian" anyway. Both are still the Sicilian. Neither is the position this article measured. That, more than any single evaluation number, is the honest summary of where this opening stands: worth playing at any rating according to some of the people who play it, including the poster who rode it from 800 to over 1400, and not worth the theory according to others in the very same threads. Our data settles which move you are likely to find. It does not settle which side of that argument is right.