Playing12 min read

Slav Defense: Humans Find 3...Nf6, Then Drift at Move 4

The Slav pages we opened are strong on variation names and global win rates. They do not say which Black move people actually choose at each rating. Chessdrive's Maia3 and Stockfish data covers two Slav positions at seven rating bands. After 3.Nf3, the engine's top move 3...Nf6 is also the human favourite at every band, rising from 32.8 percent at 1000 to 80.1 percent at 2200. After 4.Nc3 the story changes: 4...Bf5 is the likeliest move at 1000 to 1400 and costs about 0.35 against the Semi-Slav 4...e6. It is a mild inaccuracy, not a refutation, and a local Stockfish check shows why.

The starting position

The Slav Defense is 1.d4 d5 2.c4 c6, ECO codes D10 to D19, a branch of the Queen's Gambit. Black supports the d5 pawn with a pawn instead of the e-pawn, so the c8 bishop keeps its diagonal. The pages we opened explain the names well. They do not show which Black move people play at each rating, or how that compares with the engine's choice. We ran the maia3-5m checkpoint (see how Maia chess actually works) on two Slav positions at seven rating bands from 1000 to 2200, and graded every legal move with Stockfish 18 at depth 20. Percentages in the tables are the model's predicted share of human moves at that rating. They are not game database frequencies and they are not win rates.

Every evaluation is from White's side, so +0.28 means White is slightly better and -0.06 means Black is slightly better. The two dataset positions are the one after 1.d4 d5 2.c4 c6 3.Nf3 with Black to move, and the one after 3...Nf6 4.Nc3 with Black to move. Our line checks are separate local runs of Stockfish, ten seconds per line, which we call a local Stockfish check, so a number can differ by a few hundredths from the dataset. The tables are our model, and the line numbers are our engine. We never mix the two for the same claim. If you want the wider family first, read the Queen's Gambit page and the Queen's Gambit Accepted page.

Where the Slav comes from

According to Wikipedia, the Slav received an exhaustive test in the world championship matches between Alekhine and Euwe in 1935 and 1937. Euwe, Botvinnik and Smyslov particularly favoured it. A chess.com blog says 11 of the first 13 world champions played it, and that Kramnik used it in six of the eight games he had Black in the 2006 world championship. That blog is a Wikipedia derived summary, so we count it as one claim, not two.

The reputation is steady: solid, a little passive, rarely a way to lose quickly. As of October 2026, chess.com lists a free diagonal for the c8 bishop and a solid central point as pros, and slower development, a blocked c6 square for the knight and a lost tempo on ...c5 as cons. 365chess calls the Slav closed and strategic, and names a weak b7 pawn as a drawback. Our numbers say the same, and they add one thing the descriptions do not: where people go wrong.

Move 3: the human favourite is the engine move

After 3.Nf3, Stockfish ranks 3...Nf6 first at +0.28, then 3...e6 at +0.38 and 3...dxc4 at +0.43. The unusual result is that the model's favourite is the engine's favourite at every band.

Rating1000120014001600180020002200
3...Nf6 (+0.28)32.8%38.7%43.7%48.2%55.1%67.5%80.1%
3...e6 (+0.38)12.4%13.3%14.1%14.7%15.2%14.9%12.2%
3...Bf5 (+0.87)13.1%14.8%15.6%15.6%13.7%7.7%2.5%
3...Bg4 (+0.94)9.5%12.7%13.5%11.6%8.0%3.7%1.1%
3...dxc4 (+0.43)14.0%7.2%3.8%2.8%2.5%2.1%1.6%

3...Nf6 is the likeliest move at every band, from 32.8 percent at 1000 to 80.1 percent at 2200. Most openings do not behave like this: the move people play often differs from the engine's, and the gap closes only at high ratings. Here the engine move is the human move all the way down, which tells you the Slav move order is easy to find. Nobody needs a special reason to play 3...Nf6, because it is a natural developing move that also keeps every plan open. The Semi-Slav 3...e6 is a steady minority at 12.2 to 15.2 percent.

The two bishop moves are the cost. 3...Bf5 and 3...Bg4 are the develop the bishop first moves, and the dataset rates both near +0.9, the worst of the popular replies. They fade from 2000, and 3...Bg4 already from 1600. A local check of the Wikipedia warning, 3...Bf5 4.cxd5 cxd5 5.Qb3, gives +0.86 for White, and 3...Bg4 gives +0.97 with 4.Ne5 in the engine line. Both are real concessions, a bit bigger than 4...Bf5 later. The drift away from 3...dxc4 is the other change: it is 14.0 percent at 1000, but 3.8 percent at 1400 and 1.6 percent at 2200. A local check gives White +0.45 there.

Move 4: where humans and the engine split

After 3...Nf6 4.Nc3, Stockfish ranks 4...e6 first at +0.31, with 4...dxc4 and 4...a6 tied just behind at +0.34. That is a very small spread among three good moves. The human distribution is not the engine's.

Rating1000120014001600180020002200
4...e6 (+0.31)24.5%25.9%27.0%28.4%31.2%37.2%41.8%
4...Bf5 (+0.66)25.6%27.3%27.1%26.4%25.6%21.1%12.3%
4...Bg4 (+0.85)17.0%21.3%24.5%24.6%20.8%13.3%not top five
4...dxc4 (+0.34)10.4%7.0%5.3%5.4%6.9%10.9%19.3%

At 1000 to 1400 the model's favourite is 4...Bf5, at 25.6, 27.3 and 27.1 percent. From 1600, 4...e6 takes over at 28.4 percent and keeps growing to 41.8 percent at 2200. So there is a clean crossover. In this position the gap between the human favourite and the engine's best is 35 centipawns at 1000 to 1400 and zero from 1600. The first lesson in the Slav is simply to play 4...e6 or 4...dxc4 and keep the bishop home a little longer. At 2200 the picture is different again: 4...dxc4 rises to 19.3 percent, and 4...a6, the Chebanenko, shows up at 11.5 percent. Both are engine approved at +0.34.

The surprise is 4...Bg4. It is not named in the sources we opened, yet the model predicts it at 17.0 percent at 1000 and 24.6 percent at 1600. The dataset rates it +0.85, and a local check of the position after 4...Bg4 gives White +0.99 with 5.Ne5 the best move, which Black answers by retreating the bishop to c8. That is the same pattern as 4...Bf5, only larger. One caution: this table is about the immediate bishop move, so it does not describe the standard Czech line, where Black plays ...dxc4 first and ...Bf5 afterwards.

4...Bf5 is an inaccuracy, not a refutation

Two sources are harsh. Wikipedia says 4...Bf5 is not the best choice because White gains the advantage with 5.Qb3 or with 5.cxd5 and 6.Qb3. A GameKnot annotation from around 2008 calls it a common error, and says after 5.cxd5 cxd5 6.Qb3 it is extremely difficult, or to say the least impossible, for Black to defend b7 and d5. That reads like Black is losing a pawn.

The engine does not agree. In our local check, 4...Bf5 5.cxd5 cxd5 6.Qb3 is +0.62 for White, with 6...Bc8 as the best reply. The dataset puts 4...Bf5 at +0.66 against +0.31 for 4...e6, so the price is about 0.35 of a pawn. Black does not lose the b pawn. Black loses a tempo, because the bishop goes back to where it started. Wikipedia makes a similar point about 3...Bf5 4.cxd5 cxd5 5.Qb3, saying Black must retreat 5...Bc8 to save the b pawn. Our local check prefers 5...Nc6 there, at +0.86, so the retreat is not the only answer, but White's edge is larger either way. A GameKnot commenter prefers 6...Qb6 to 6...Bc8 because the retreat loses time to Bf4. We did not get an engine ruling on that idea, so we treat it as an opinion.

This matters for practical play. A third of a pawn is something a 1200 player will not convert reliably. So if you like the bishop on f5, learn the real version. In the standard Czech main line, Black first takes on c4, then plays ...Bf5 after 5.a4. Wikipedia calls this the main line of the Slav, and says 5...Bf5 prevents 6.e4. A local check of the position after 5.a4 gives +0.39 with 5...e6 as the best move, and the full main line 6.e3 e6 7.Bxc4 Bb4 8.O-O is +0.23. That is a smaller concession than the immediate bishop move.

The Semi-Slav: the engine's choice and the sharper one

Playing 4...e6 turns the opening into the Semi-Slav, ECO D43 to D49. Wikipedia calls it a very complex opening in its own right, and says about 80 percent of games continue 5.Bg5 or 5.e3. The chess.com forum thread on Slav or Semi-Slav is mostly opinion, but the opinions agree: one poster says the Semi-Slav is more popular and sharper in lines like the Botvinnik, Noteboom and Marshall Gambit, another that the Slav is more solid. We checked the main branches.

  • Moscow, 5.Bg5 h6. A local check gives +0.35, with 6.Bh4 the engine move. Wikipedia lists this inside the Anti-Meran Gambit, where Black can also play the Cambridge Springs 5...Nbd7, the Botvinnik 5...dxc4 or the Orthodox 5...Be7.
  • Botvinnik, 5.Bg5 dxc4 6.e4 b5. The sharpest option, at +0.57 in our check, with 7.e5 the engine move.
  • Meran, 5.e3 Nbd7 6.Bd3 dxc4 7.Bxc4 b5. +0.22 at depth 31, the smallest White edge of the three, though not a quiet line. The first Meran game on Wikipedia's list is Schlechter against Perlis in 1906, and the Reynolds Attack 10.e5 is called wildly tactical.
  • Exchange, 5.cxd5 exd5. Wikipedia says this should give equality, and it is the safest way to get a quiet game if White wants one.

The sidelines, checked

Most Slav pages list a few named alternatives to the main line. We checked the ones the sources name.

The Geller Gambit, 5.e4

After 4...dxc4, White can play 5.e4 b5. Wikipedia said in 2005 that this is generally thought to favour Black. Our local check gives -0.06 from White's side with 6.Be2 the engine move, so it is about equal, not better for Black. Black is fine, and so is White. It is a sharp line that does not decide the game.

The Chebanenko, 4...a6

The Chebanenko is described by publishers as an extremely flexible and almost universal weapon against the Queen's Gambit, and it is reportedly popular with Bologan and Shirov. As of October 2026, chess.com's database says it is the second most popular choice after 4...dxc4. The dataset gives it +0.34, tied for second best, but only 2200 players choose it often enough to appear in the top five, at 11.5 percent. A local check after 4...a6 gives +0.50 with 5.c5 as White's best, so White gets a small space edge.

The Exchange Slav, 3.cxd5

After 3.cxd5 cxd5, Wikipedia calls the position drawish, 365chess calls it the most drawish of the Slav variations, and as of October 2026 chess.com reports more than 4,600 games with 65 percent drawn. As of October 2026, 365chess shows 3.cxd5 at 24.8, 54.3 and 20.9 percent across 11,962 games, reading its unlabelled columns as White win, draw and Black win, which is the lowest White score of any third move apart from rare ones. A local check gives +0.32, so it is a real offer of a quiet game, not a trick.

What the global statistics say

Here the sources disagree, so we label each one. As of October 2026, chess.com reports that 2...c6 is the most popular reply to 2.c4, and on its Slav page gives White about 40 percent, draws about 39 percent and Black about 21 percent. Our earlier page on the Queen's Gambit quotes a different chess.com snapshot from September 2026: 118,063 games, with White at 40.0 percent, draws at 37.2 percent and Black at 22.8 percent. As of October 2026, 365chess lists 102,039 Slav games at 37.8, 39.7 and 22.5 percent, with unlabelled columns read as White, draw and Black. These are three snapshots of different pools and dates, so do not average them. The shared message is simple: White scores a little under 40 percent, and the Slav draws a lot.

Our data adds the missing layer: none of those pages say which move people play, or that the human favourite at 1000 to 1400 after 4.Nc3 is a mild inaccuracy. Maia3 is a model of human choice trained on online games, not a count of real games, so read every percentage above as a prediction.

Who should play the Slav, by rating

Our data does not include game results by rating for the Slav, so this is what the model and the sources suggest, not a result table.

  • Around 1000 to 1400. You are the group most likely to meet and play 4...Bf5, at 25.6 to 27.3 percent. Play 4...e6 or 4...dxc4 instead, and keep Bf5 for after you have taken on c4. A forum poster recommends the Slav especially at the 1300 to 1400 level, even though White can steer toward a draw with the Exchange.
  • Around 1600 to 1800. 4...e6 becomes the likeliest move at 28.4 to 31.2 percent, so this is where players start to choose the Semi-Slav.
  • 2000 and above. 3...Nf6 is 67.5 to 80.1 percent, 4...dxc4 grows to 19.3 percent at 2200 and the Chebanenko appears. This is where memorised theory matters, and the argument in why you should stop memorizing openings and learn plans only goes so far.

If you want a different pairing against 1.d4, the London System page covers a White system that skips 2.c4 altogether, and the King's Indian Defense page is the other major answer, with much sharper play. For solid answers to 1.e4, the Caro Kann Defense page has the same pawn on c6 idea. For how all of this fits with other openings by strength, see chess openings by rating.

Where other resources are better

This page is narrow on purpose. The dataset holds two positions, and Maia3 5m is a small model of human choice, so its percentages are predictions and can be wrong at the margins. Our line checks are quick local runs, not a substitute for an opening database or a repertoire course. Wikipedia is the best reference for the full list of named variations, including the Czech, Dutch, Lasker and Krause lines, and for the history. As of October 2026, chess.com has a clear list of pros and cons with win rates for six named lines. 365chess has game counts for each third move, though its result columns are not labelled. We have no game results by rating, and we did not open the Lichess explorer, which is the place for real game counts. We also found no sourced elite Slav game from 2024 or 2025, and no traps named by chess.com. None of the pages we opened compares the engine's best move with the likeliest human move at each rating, which is the gap this one fills.

What to take from it

The Slav is easy to start. After 3.Nf3, the model predicts the engine move 3...Nf6 at every band, from 32.8 percent at 1000 to 80.1 percent at 2200. The decision that separates ratings comes one move later: after 4.Nc3 the human favourite at 1000 to 1400 is 4...Bf5, which costs about 0.35 against 4...e6, and 4...e6 becomes the favourite from 1600. Call that a mild inaccuracy, +0.62 in a local check after 5.cxd5 cxd5 6.Qb3 Bc8, not a refutation and not a lost pawn. The Semi-Slav lines we checked sit between +0.22 and +0.57. Play 4...e6 if you want the engine's choice, 4...dxc4 if you want the classical route, and keep every percentage in this article for what it is: the prediction of a model, not a count of real games.

Frequently asked

Is the Slav Defense good?
Yes, it is sound. In our Stockfish data the Slav main move 3...Nf6 is the engine's top reply after 3.Nf3 at +0.28 from White's side, and after 4.Nc3 the best move is 4...e6 at +0.31. As of October 2026, chess.com lists a free diagonal for the c8 bishop and a solid central point as pros, and slower development as a con. White keeps a small edge, which is normal for the second player, in nearly every line we checked. The exception is the Geller Gambit at -0.06.
Is the Slav Defense good for beginners?
It is a fair choice, because the structure is solid and the ideas repeat. The trap for beginners is moving the c8 bishop too soon. In the dataset 4...Bf5 is the single most likely move at 1000 to 1400, at 25.6 to 27.3 percent, and it is slightly worse than 4...e6. It is not losing, so you can learn from it without being punished hard.
What is the difference between the Slav and the Semi-Slav?
The Slav is 1.d4 d5 2.c4 c6 with the knight coming to f6 and the bishop to f5 or g4. The Semi-Slav adds ...e6, so after 3.Nf3 Nf6 4.Nc3 it is 4...e6. The bishop on c8 is locked in for a while, in exchange for a very solid centre. Wikipedia calls it a very complex opening in its own right. Our engine ranks 4...e6 first at +0.31 and humans choose it most from 1600 up.
Is 4...Bf5 a mistake in the Slav?
A small one. Wikipedia says it is not the best choice and a GameKnot annotator calls it a common error. The dataset gives it +0.66 against +0.31 for 4...e6, and a local ten second Stockfish check of 5.cxd5 cxd5 6.Qb3 Bc8 gives +0.62. That is a concession of roughly a third of a pawn. Black does not lose the b pawn if the bishop retreats.
Is the Exchange Slav drawish?
Yes, by every source we opened. Wikipedia calls it drawish, 365chess calls it the most drawish of the Slav variations, and as of October 2026 chess.com reports more than 4,600 games with 65 percent drawn. A local check of 3.cxd5 cxd5 gives +0.32, so White has little more than the move.
Which world champions played the Slav?
A chess.com blog says it was played by 11 of the first 13 world champions, and Wikipedia says it received an exhaustive test in the Alekhine and Euwe matches of 1935 and 1937. Euwe, Botvinnik and Smyslov favoured it. We found no sourced elite Slav game from 2024 or 2025, so we do not claim a current trend.

Want to play the Slav against a model that answers like 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 the Slav from either side against a model that picks its moves the way your band does, then see where the game turned with a separate Stockfish review. Available now on iOS.

Keep reading

All guides

Updated