Open with 1.b3 and you tell the opponent that you want a bishop on the long diagonal and a game with less theory. That is the Nimzowitsch-Larsen Attack, also called Larsen's Opening or the Queen's Fianchetto Opening. Wikipedia gives it ECO codes A01 to A06, depending on move order. This page asks a narrow question: what do players actually play after 1.b3 at each rating, and how does that compare with the engine's line?
The pages we opened cover names, history and a few lines. Only Chessiverse shows reply shares by rating, and only at 1200 and 2500. We ran the maia3-5m checkpoint (see how Maia chess actually works) on two positions at seven rating bands from 1000 to 2200 and graded every legal move with Stockfish 18 at depth 20. The percentages below are the model's predicted move probabilities for a player at that rating. They are not game database counts, and they are not win rates.
Every table evaluation is from White's side, so -0.14 means Black is better by 0.14 pawns. The two positions are the one after 1.b3 and the one after 1.b3 e5 2.Bb2, both with Black to move. The second is the position the model reaches by following its own favourite moves at 1400. Our line checks are separate local Stockfish runs, called local checks below, and their evaluations are given from the side named in the sentence.
Where 1.b3 comes from
Wikipedia says John Owen occasionally opened 1.b3 in the 1870s, essentially playing Owen's Defence, 1.e4 b6, with a move in hand. Aron Nimzowitsch experimented with 1.b3 in the 1920s and 1930s but more often played 1.Nf3 d5 2.b3. Bent Larsen played 1.b3 frequently from 1968 to 1972, inspired by Nimzowitsch's setup, which is how the combined name arose. Chess.com adds that the earliest recorded serious 1.b3 game dates from 1851.
The ordering matters. 1.Nf3 first stops 1...e5, so the Nimzowitsch route keeps Black out of the open centre. Larsen's 1.b3 allows 1...e5 and invites the fight over that pawn, which is exactly what makes the early data interesting.
Later players kept it going. Wikipedia says Vladimir Bagirov played 1.b3 about 100 times between 1976 and 2000, scoring roughly 54 percent wins and 38 percent draws. Chess.com says Fischer beat Mecking with it at the 1970 Palma de Mallorca Interzonal. As of October 2026, chess.com lists 1.b3 as the sixth most popular first move in its master database, behind 1.e4, 1.d4, 1.c4, 1.Nf3 and 1.g3.
Move 1: the engine's reply and the common one are the same until 1800
After 1.b3, Stockfish ranks 1...e5 first at -0.14, then 1...d5 at -0.03 and 1...Nf6 at +0.05. Sixteen of Black's 20 replies sit between -0.14 and +0.43. This is a quiet first move, not a mistake, so every sensible reply is close to equal.
| Rating | 1000 | 1200 | 1400 | 1600 | 1800 | 2000 | 2200 |
|---|---|---|---|---|---|---|---|
| 1...e5 (-0.14) | 50.3% | 47.9% | 43.1% | 37.0% | 30.6% | 25.2% | 23.0% |
| 1...d5 (-0.03) | 22.6% | 23.2% | 24.1% | 25.4% | 26.4% | 25.8% | 23.9% |
| 1...Nf6 (+0.05) | 4.4% | 5.0% | 6.2% | 8.2% | 11.3% | 16.6% | 21.8% |
| 1...e6 (+0.13) | 5.1% | 5.8% | 6.7% | 7.2% | 7.3% | 6.8% | 5.4% |
| 1...c5 (+0.12) | not top five | not top five | 4.5% | 5.7% | 6.8% | 7.4% | 7.3% |
The shape is clear. 1...e5 is the favourite from 1000 to 1800, then falls from 50.3 percent at 1000 to 23.0 percent at 2200. At 2000 and 2200 the likeliest reply flips to 1...d5, at 25.8 and 23.9 percent. Meanwhile 1...Nf6 climbs from 4.4 percent to 21.8 percent.
Be careful with what that says. It does not mean weaker players find the best move more often in any meaningful sense. The engine's first choice is only 11 centipawns ahead of 1...d5 and 19 ahead of 1...Nf6, which is noise in an opening position. Stronger players spread their replies across several near-equal moves, so the share of any single move shrinks. At 2200, 1...d5, 1...e5 and 1...Nf6 are predicted within 2.1 percentage points of each other. The weaker bands simply funnel into the most natural central move.
The one banded source we found agrees in shape. As of October 2026, Chessiverse reports Lichess reply shares of 46.3 percent for 1...e5, 23.9 percent for 1...d5 and 6.4 percent for 1...e6 at 1200, against our modelled 47.9, 23.2 and 5.8 percent. At 2500 it shows 1...e5 at 28.3 percent, 1...d5 at 26.5 percent and 1...Nf6 at 20.2 percent, which sits where our 2200 column is heading. Its overall figures are about 45 percent for 1...e5, 27 percent for 1...d5 and 15 percent for 1...Nf6. Chessiverse and Wikipedia call 1...e5 the Modern Variation and 1...d5 the Classical Variation, and 1...Nf6 is the Indian Variation. Treat "main reply" as a statement about a band, not about the opening.
Move 2: does Black have to defend e5?
The model follows 1...e5 with 2.Bb2 at 92.3 percent at 1400, so 1.b3 e5 2.Bb2 is the natural second position. White's bishop eyes the e5 pawn, and chess.com says Black can defend it easily with 2...d6 or 2...Nc6, with 2...Nc6 a bit more common. Our data says the same, with numbers.
| Rating | 1000 | 1200 | 1400 | 1600 | 1800 | 2000 | 2200 |
|---|---|---|---|---|---|---|---|
| 2...Nc6 (-0.03) | 57.6% | 63.6% | 64.2% | 61.9% | 59.7% | 61.4% | 67.5% |
| 2...d6 (+0.13) | 11.5% | 12.3% | 15.1% | 18.9% | 22.9% | 25.2% | 23.9% |
| 2...f6 (+0.53) | 7.4% | 5.6% | 4.7% | 4.6% | 5.0% | 5.3% | 4.6% |
| 2...Qf6 (+0.69) | 4.9% | 3.7% | not top five | not top five | not top five | not top five | not top five |
| 2...d5 (+0.95) | 4.7% | 3.9% | 3.6% | 3.5% | 3.0% | 1.9% | 0.8% |
2...Nc6 is both the engine's best move and the likeliest move at every band, between 57.6 and 67.5 percent. 2...d6, which is 16 centipawns worse, grows from 11.5 percent at 1000 to 25.2 percent at 2000. In other words, Black almost always defends e5 with a developing move, and the data says what chess.com says: the pawn is not under real pressure.
The pitfall: 2...d5 and 2...Nf6 leave e5 hanging
No source we read says this directly, so the check is ours. After 1.b3 e5 2.Bb2 d5, White plays 3.Bxe5 and a local check puts White ahead by about 1.00. After 2...Nf6, 3.Bxe5 is again the top move, again about +1.00 for White. The local check's best defence in either line still leaves White a pawn up.
How often does it happen? Rarely. 2...d5 is predicted at 4.7 percent at 1000, 3.9 at 1200, 3.6 at 1400, 3.5 at 1600, 3.0 at 1800, 1.9 at 2000 and 0.8 percent at 2200. It is a small pitfall that fades with rating, not a trap that wins games on its own. If you want to play for it as White, 1.b3 e5 2.Bb2 offers it, but you will mostly see 2...Nc6.
The more common lower-band slips are quieter. 2...f6, at +0.53 for White, is predicted at 7.4 percent at 1000 and still 4.6 percent at 2200. The engine says it gives White about half a pawn. 2...Qf6 at +0.69 appears at 4.9 percent at 1000 and 3.7 percent at 1200 and drops out of the top five after that. Compare that to the 2...d5 reply, and you can see why a White player at 1000 should expect an inexact move more often than a hanging pawn.
The Indian line trap
Chessiverse lists a trap against the Indian Variation: 1.b3 Nf6 2.Bb2 g6 3.e4, attacking the knight. If Black plays 3...Nxe4??, then 4.Bxh8 wins a rook, because the long diagonal is open all the way. We did not run the Indian line in our dataset, and the source gives no frequency, so we cannot say how often 3...Nxe4 appears. The pattern is worth knowing for both sides: after ...g6 the g7 square is empty, so once the knight on f6 leaves, the bishop on b2 hits the rook on h8.
Larsen against Spassky: the famous loss, move by move
Larsen's best-known 1.b3 game is a loss. The score below is from ChessBase and matches an independent LaTeX example file, and chess.com and Wikipedia both agree on the 17-move length. It was played in Belgrade, in the USSR against Rest of the World match, in March 1970:
1.b3 e5 2.Bb2 Nc6 3.c4 Nf6 4.Nf3 e4 5.Nd4 Bc5 6.Nxc6 dxc6 7.e3 Bf5 8.Qc2 Qe7 9.Be2 O-O-O 10.f4 Ng4 11.g3 h5 12.h3 h4 13.hxg4 hxg3 14.Rg1 Rh1 15.Rxh1 g2 16.Rf1 Qh4+ 17.Kd1 gxf1=Q+ 0-1
We replayed it and every move is legal. The question is where White first went wrong, and the engine's answer is later than you might expect.
- 4.Nf3 is not the mistake. After 3...Nf6 a local check says 4.Nf3 is Stockfish's top move, at -0.08 from White's side. The reply 4...e4 is then Black's best at +0.30 for Black, so the position is balanced and Black has only a tiny edge.
- After 9...O-O-O Black is better. White's best move at that point is 10.a3, and Stockfish still gives -0.75 from White's side. So by move 9 White's position has already slipped about three quarters of a pawn.
- 10.f4 is where the game turns. The game move is not in the local check's top three. After 10.f4 the evaluation drops to about -2.8 from White's side, and 10...Ng4 starts Black's kingside play. By the time Black plays 12...h4, White is already about five pawns down on the engine's count.
- The finish. After 13.hxg4 hxg3, Stockfish gives Black about +5.4, against only +0.34 for 13...Bxe3. A check at lower depth puts 14...Rh1 after 14.Rg1 at about +5.7. The famous rook sacrifice is the finish of a won game, not the moment it was won.
That is a useful lesson for anyone studying the opening. Chess.com calls the game a 17-move brilliancy with a spectacular 14th move, and it is. But the opening itself, 1.b3 e5 2.Bb2 Nc6 3.c4 Nf6, was fine for White. Larsen lost it in the middlegame with 10.f4, which walked into a kingside pawn storm, not by playing 1.b3.
The main lines in print
Wikipedia gives two main ways to play after 1.b3 e5 2.Bb2 Nc6. The first is 3.e3 d5 4.Bb5 Bd6 5.f4, where 4.Bb5 threatens Bxc6+ and Bxe5, and the source says that after 5.f4 Black cannot take on f4 with the e5 pawn because Bxg7 wins the h8 rook. We did not check that line, so we repeat it as sourced. The second is 3.c4, which Wikipedia compares to a reversed Sicilian: 3...Nf6 4.e3 d5 5.cxd5 Nxd5. Larsen's game against Spassky took the 3.c4 route.
Against 1...d5, Wikipedia gives 2.Bb2 c5 3.e3 Nc6 4.Nf3 Nf6 5.Bb5, which has a reversed Nimzo-Indian feel, and 2.Bb2 Bf5 3.d3 e6 4.Nd2 Nf6 5.g3 Be7 6.Bg2 h6 7.e4 Bh7, a Hippopotamus-type structure. Chess.com says that after 1...d5 and 1...Nf6, Black usually plays the other of the two next, so no pawn is offered to the bishop. Our dataset stops at move 2 after 1...e5, so none of those lines carries a number from us.
Who should play it, by rating
Our data has no game results by rating, so this is what the model and the sources suggest, not a result table.
- Around 1000 to 1400, as White. Half of all replies at 1000 are 1...e5, and 2...Nc6 follows over half the time. Once 1.b3 e5 2.Bb2 is on the board, the inexact 2...f6 is about one reply in 14 at 1000. The e5 pitfall is real but rare, so do not build your plan around it.
- Around 1600 to 1800, as White. 1...d5 climbs to 26.4 percent and 1...Nf6 to 11.3 percent at 1800. You will meet the Classical and Indian lines more, and after 2.Bb2 Black's 2...d6 is roughly one reply in five.
- 2000 and above. The replies spread out. 1...d5, 1...e5 and 1...Nf6 are all common, and 2...Nc6 remains the answer to 2.Bb2 six times in ten. Preparing for one reply is not enough.
- As Black, at any rating. 1...e5 is the engine's choice, and 2...Nc6 is the answer to 2.Bb2. Skip 2...d5 and 2...Nf6, which give up the pawn.
Duolingo calls the opening sound but rare at the top levels, and says it usually leads to solid, positional games. Chess.com lists long-diagonal pressure and ease of learning as pros, and slower kingside development and a bishop that can be shut out as cons. Our data is consistent with sound: Black is never better than 0.14 pawns after 1.b3. For the neighbouring flank opening that gets a far worse engine verdict, see the Grob Opening page. For queen's pawn setups that also dodge theory, the London System page and the Dutch Defense page are the next stops, and chess openings by rating shows how all of this fits together. If you would rather spend the time on plans than move orders, stop memorizing openings and learn plans argues the case.
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 not a count of real games. Our line checks are quick local runs, not a repertoire. Wikipedia is better for the history and the full game scores, chess.com for a plain tour of the structure, and Chessiverse for Lichess win rates and the Indian line trap. We did not open the Lichess explorer, which is the place for real game counts, and we did not obtain the Fischer and Mecking move list. None of the pages we read compares the engine's best reply with the likeliest human reply at each rating, which is the gap this one fills.
What to take from it
After 1.b3, 1...e5 is the engine's best reply by a small margin and the likeliest one up to 1800, predicted at 50.3 percent at 1000 and 23.0 percent at 2200, where 1...d5 narrowly leads. Stronger players spread across near-equal moves, so do not read that fall as a skill signal. After 2.Bb2, 2...Nc6 is best and likeliest at every band, and the e5 pitfall with 2...d5 or 2...Nf6 costs about a pawn but is rare. Larsen's loss to Spassky began with a fine opening and turned on 10.f4, about two pawns of evaluation in one move. Keep every percentage for what it is: the prediction of a model, not a count of real games.
Frequently asked
- What is the Nimzowitsch-Larsen Attack?
- It is the flank opening 1.b3, where White plans a bishop on b2 aimed down the long diagonal. Wikipedia gives it ECO codes A01 to A06 (A01 for independent 1.b3 lines, A04 to A06 for 1.Nf3 move orders with 2.b3) and lists the alternative names Larsen's Opening and the Queen's Fianchetto Opening.
- What is the best reply to 1.b3?
- In our Stockfish data 1...e5 is first at -0.14 from White's side, with 1...d5 at -0.03 and 1...Nf6 at +0.05. The gap between them is tiny, so all three are fine. After 1...e5 2.Bb2 the best move is 2...Nc6, with 2...d6 at +0.13 a close second.
- Does 2.Bb2 win the e5 pawn?
- Only if Black ignores it. After 1.b3 e5 2.Bb2, a local Stockfish check shows that 2...d5 or 2...Nf6 allows 3.Bxe5, worth about +1.00 for White. Our model predicts those moves rarely: 2...d5 at 4.7 percent at 1000, falling to 0.8 percent at 2200. 2...Nc6 and 2...d6 defend the pawn while developing.
- Who invented 1.b3?
- Nobody single-handedly. Wikipedia says John Owen opened 1.b3 occasionally in the 1870s, Aron Nimzowitsch experimented with it in the 1920s and 1930s but more often played 1.Nf3 d5 2.b3, and Bent Larsen played it frequently from 1968 to 1972. The Nimzowitsch-Larsen name joins Nimzowitsch's 1.Nf3 and 2.b3 setup to Larsen's 1.b3.
- Did Fischer play the Nimzo-Larsen?
- Yes. He played 1.b3 in 1970, including a win over Mecking at the Palma de Mallorca Interzonal, according to chess.com. We did not obtain the move list, so we do not analyse that game.
- How did Larsen lose to Spassky in 17 moves?
- In Belgrade in March 1970, Larsen played 1.b3 e5 2.Bb2 Nc6 3.c4 Nf6 4.Nf3 e4 and lost after a kingside pawn storm ended in 17...gxf1=Q+. Our check shows 4.Nf3 is Stockfish's top move and Black is already better after 9...O-O-O, at -0.75 from White's side. The game turns on 10.f4, after which the evaluation drops to about -2.8, and by the time of 13.hxg4 hxg3 Stockfish gives Black about +5.4.
Want to face 1.b3 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 Nimzo-Larsen 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
Updated