The best first move in chess is a three-way tie. White has exactly 20 legal first moves, sixteen pawn moves and four knight moves, and after one move each there are 400 possible positions. Most first moves are fine. Three are tied at the top by the strongest engine we can run, and a handful are bad. We scored all 20 with Stockfish 18 at depth 24, then asked Maia3 5m, a neural model trained on human games, what players at each rating tend to choose.
The short answer
1.e4, 1.d4 and 1.Nf3 are tied. Stockfish 18 at depth 24 gives 1.e4 +0.32, 1.d4 +0.32 and 1.Nf3 +0.30. Next comes 1.c4 at +0.17. A gap of 0.02 pawns between the top three is noise for any human game, and no first move wins by force. A deeper engine run can reorder them, as one public depth-78 Stockfish 16 run does, which is why we call it a tie and not a ranking.
So pick by what you will understand, not by the score. If you are new, play 1.e4. If you like slower positions, play 1.d4. The sections below give the full table, what players actually choose, and why Black's reply matters more than your first move.
All 20 first moves, ranked by Stockfish 18
Evals are from White's point of view, in pawns, at depth 24. A positive number favours White.
| Rank | First move | Eval | Rank | First move | Eval |
|---|---|---|---|---|---|
| 1 | 1.e4 | +0.32 | 11 | 1.b3 | -0.09 |
| 2 | 1.d4 | +0.32 | 12 | 1.h3 | -0.16 |
| 3 | 1.Nf3 | +0.30 | 13 | 1.a4 | -0.16 |
| 4 | 1.c4 | +0.17 | 14 | 1.b4 | -0.18 |
| 5 | 1.g3 | +0.17 | 15 | 1.f4 | -0.27 |
| 6 | 1.e3 | +0.16 | 16 | 1.h4 | -0.49 |
| 7 | 1.c3 | +0.07 | 17 | 1.Nh3 | -0.62 |
| 8 | 1.Nc3 | +0.03 | 18 | 1.Na3 | -0.66 |
| 9 | 1.a3 | +0.02 | 19 | 1.f3 | -0.81 |
| 10 | 1.d3 | -0.04 | 20 | 1.g4 | -1.26 |
Chess.com's own ranking article covers all 20 first moves too, but its tiers are an author's opinion with no engine numbers. Ours are measured. Two things stand out. First, the playable zone is wider than most beginners think: the top six moves sit within 0.16 pawns of each other, and 1.c4, 1.g3 and 1.e3 are all fine. Second, only the bottom few are really bad. 1.g4 at -1.26 is the worst, and 1.f3 at -0.81 is the first move of Fool's Mate (1.f3 e5 2.g4 Qh4#), the shortest possible checkmate at 4 plies. That tier list agrees on the shape of this, putting 1.f3, 1.a4, 1.h4 and 1.h3 at the bottom and 1.e4 and 1.d4 at the top, though that ranking is the author's judgment and not data.
What players actually play, by rating
Engines say what is strong. They do not say what people choose. For that we used Maia3 5m, which predicts the move a player at a given rating would make. The figures below are the model's predicted probabilities for White's first move. They are model output, not game counts, so read them as "Maia predicts a player at 1200 opens 1.e4 about this often", not as a head count.
| Rating | 1.e4 | 1.d4 | 1.Nf3 | 1.c4 | 1.e3 |
|---|---|---|---|---|---|
| 600 | 62.9% | 21.5% | 2.5% | 1.5% | 3.1% |
| 1000 | 67.8% | 21.3% | 2.1% | 1.4% | 2.5% |
| 1400 | 65.3% | 23.5% | 2.2% | 2.2% | 1.8% |
| 1800 | 60.7% | 26.2% | 2.8% | 3.5% | 1.2% |
| 2200 | 50.8% | 30.9% | 6.4% | 5.8% | 0.7% |
| 2400 | 46.5% | 33.9% | 8.2% | 6.0% | 0.5% |
The pattern is clear. 1.e4 peaks around 67.8% at 1000 (it is 67.3% at 800) and falls to 46.5% at 2400. 1.d4 climbs steadily from 21.3% to 33.9%. 1.Nf3 and 1.c4 stay small until about 2000, where they reach 4.0% and 4.5%, then rise to 8.2% and 6.0% at 2400. The odd 1.e3, a quiet move the engine rates at +0.16, is a lower-rated habit: 3.1% at 600 and 0.5% at 2400.
Does the model match real games?
It should, if the model is any good, so we checked against real Lichess blitz and rapid shares published by Chessiverse (sample size and date are not stated on their page, so treat their figures as indicative). They report, for White's first move, 1.e4 65.8%, 1.d4 20.2% and 1.Nf3 2.7% at rating 400, and 1.e4 44.6%, 1.d4 34.5% and 1.Nf3 10.4% at rating 2500.
| Source | Rating | 1.e4 | 1.d4 | 1.Nf3 |
|---|---|---|---|---|
| Lichess shares (Chessiverse) | 400 | 65.8% | 20.2% | 2.7% |
| Maia3 5m prediction | 600 | 62.9% | 21.5% | 2.5% |
| Lichess shares (Chessiverse) | 2500 | 44.6% | 34.5% | 10.4% |
| Maia3 5m prediction | 2400 | 46.5% | 33.9% | 8.2% |
The two agree at both ends. The bands are not identical (400 against 600, 2500 against 2400), and the model is a few points low on 1.e4 at the bottom and a couple of points low on 1.Nf3 at the top. But the story is the same: a heavy lean toward 1.e4 for beginners, then a gradual move to 1.d4, 1.Nf3 and 1.c4 as players improve. We found real shares only for those two ratings and no 1.c4 figure, so the middle of our table rests on the model alone.
Black's reply matters more than your first move
Choosing among the top three first moves changes the engine score by 0.02. Black's answer can change it by half a pawn. We scored every Black reply after 1.e4, 1.d4, 1.c4 and 1.Nf3, and asked Maia3 what each rating plays.
After 1.e4
Stockfish puts ...e5 first at +0.25. Maia3 agrees on what low-rated players play, with ...e5 at 65.8% at 600, but ...d5 is the second choice at 10.1% and the engine scores it at +0.71, nearly half a pawn worse for Black. The Sicilian ...c5 (+0.37) takes over from ...e5 at 2000, where Maia3 gives it 27.8% against 26.5%.
| Rating | ...e5 | ...c5 | ...e6 | ...c6 | ...d5 |
|---|---|---|---|---|---|
| 600 | 65.8% | 3.1% | 3.6% | 1.7% | 10.1% |
| 1000 | 62.3% | 6.1% | 5.4% | 4.3% | 11.8% |
| 1400 | 52.2% | 13.0% | 8.1% | 6.6% | 9.4% |
| 1800 | 35.6% | 22.3% | 11.0% | 8.5% | 9.7% |
| 2000 | 26.5% | 27.8% | 12.5% | 9.9% | 8.9% |
| 2400 | 17.1% | 37.5% | 12.6% | 12.9% | 4.8% |
| Stockfish eval | +0.25 | +0.37 | +0.39 | +0.29 | +0.71 |
This is why 1.e4 is a friendly choice for a beginner. Against players below 1200, the model expects ...e5 most of the time, which gives you the open games covered in our guide to the kings pawn opening. A March 2025 analysis of the Lichess database by Chess Engine Lab finds the same pattern in real games: at lower levels White scores much better against 1.e4 e5 than against other common replies, and Black's replies converge at higher ratings.
After 1.d4
The engine's best reply is ...Nf6 at +0.26, with ...d5 at +0.29. Maia3 predicts ...d5 dominates until the middle ratings: 50.4% at 600 and 56.0% at 1000. ...Nf6 overtakes it at 2200, with 40.5% against 28.0%. The reply the model gives at 600 as its second choice, ...e5 at 16.9%, scores +1.39 for White, because it simply offers a pawn.
| Rating | ...d5 | ...Nf6 | ...e6 | ...e5 |
|---|---|---|---|---|
| 600 | 50.4% | 4.0% | 7.6% | 16.9% |
| 1000 | 56.0% | 8.0% | 7.9% | 11.2% |
| 1400 | 49.7% | 12.8% | 9.4% | 9.2% |
| 1800 | 41.3% | 21.2% | 9.7% | 6.4% |
| 2200 | 28.0% | 40.5% | 7.8% | 2.1% |
| 2400 | 26.2% | 47.9% | 6.1% | 0.8% |
The pawn offer fades as ratings rise, to 2.1% at 2200 and 0.8% at 2400. Either way, a player who opens 1.d4 meets ...d5 first, so the queen's pawn games in our queens pawn opening guide are the ones to learn.
After 1.Nf3 and 1.c4
Here the distance between engine and model is largest. After 1.Nf3, Stockfish rates ...d5, ...Nf6 and ...e6 all at +0.27, but ...e5 at +1.34, because the pawn falls to 2.Nxe5. Maia3 still predicts ...e5 at 26.0% at 600, 16.8% at 800 and 10.0% at 1000. A player at 600 who answers 1.Nf3 as if it were 1.e4 drops about a pawn by the engine's numbers, so the reply you meet matters even when your own move is sound.
| Rating | ...d5 | ...Nf6 | ...e5 | ...c5 |
|---|---|---|---|---|
| 600 | 29.9% | 10.3% | 26.0% | 2.9% |
| 1000 | 41.3% | 7.1% | 10.0% | 3.8% |
| 1800 | 35.6% | 13.9% | 3.3% | 10.4% |
| 2200 | 29.2% | 29.3% | 1.0% | 11.8% |
| 2400 | 29.1% | 35.2% | 0.4% | 10.8% |
| Stockfish eval | +0.27 | +0.27 | +1.34 | +0.42 |
After 1.c4, Stockfish scores ...e5 at +0.26, ...e6 at +0.29, ...Nf6 at +0.31 and ...c5 at +0.32, all inside a few hundredths. Maia3 predicts ...e5 at 48.2% at 600 and 50.9% at 1000, and ...Nf6 pulls level at 2000 (24.2% against 23.9%) before ...e5 slips to 20.4% at 2400. One public Stockfish 16 run reads 1.c4 as heading toward zero at great depth, and at depth 24 we already score it below the top three.
A recommendation by level
All three of the top moves follow the same opening ideas: take space in the center, free your pieces, and get your king safe. The first move only chooses which of those ideas you start with. The other 17 moves are not all errors, but they either give up central control or, like 1.g4 and 1.f3, weaken your own king.
Beginners: 1.e4. It opens lines for the queen and bishop, it invites quick development, and our model data says players below 1200 mostly answer ...e5, so the positions repeat. Start with our guide on how to play chess for beginners, then learn the main 1.e4 e5 lines.
If you like structure: 1.d4. The engine scores it identically, the games are usually slower, and you get plans built around pawn breaks and piece placement. It is the more common choice as players improve, rising from 21.3% at 1000 to 33.9% at 2400 in the model.
If you like flexibility: 1.Nf3 or 1.c4. Both keep your pawn structure undecided and can transpose into many setups. The cost is that you must be ready for more kinds of reply. For how opening choice shifts across ratings, see our breakdown of chess openings by rating.
For history: Bobby Fischer called 1.e4 "best by test" in Chess Life in December 1963, according to the chess historian Edward Winter, in his report of a win over Bisguier. Treat that as history, not proof. It was a human judgment made long before engines, and the engine now says 1.d4 is its equal.
What this data cannot tell you
- It has no win rates. We do not have win, draw and loss results by first move and rating. Public explorer win rates by first move are hard to read, because the rating filter and transpositions blur them, so we did not quote any. Public tables disagree on whether 1.d4, 1.c4 or 1.Nf3 scores best.
- Maia3 is a model. It predicts a move for a given rating. It is not a game count. We checked it against real shares at two ratings only.
- Depth and engine matter. Our scores are Stockfish 18 at depth 24. Other depths can reorder the top three, so the tie is the finding, not the order.
- An engine score is not a plan. A +0.32 position is only as good as your understanding of it. For more on that limit, read our piece on best chess moves and what a calculator won't tell you.
Method
On 2026-11-13 we ran Stockfish 18 at depth 24 on the start position, scoring all 20 legal first moves, and on the position after each of 1.e4, 1.d4, 1.c4 and 1.Nf3, scoring every legal Black reply. Evals are White's point of view in pawns. We ran Maia3 5m at self ratings from 600 to 2400 and read its full move distribution, rounding probabilities to one decimal place. Maia3 probabilities are what the model assigns to each move, not counts of games, and every number on this page comes from those two runs or from the cited Chessiverse page. Real-game shares come from the Chessiverse move trends page, which draws on the Lichess opening explorer for blitz and rapid games.
Frequently asked
- What is the best first move in chess?
- There is no single best one. Stockfish 18 at depth 24 scores 1.e4 and 1.d4 at +0.32 and 1.Nf3 at +0.30, a gap of 0.02 pawns, and 1.c4 at +0.17. For most players the best first move is the one whose positions you understand, and 1.e4 is a good default for beginners.
- What are the best starting moves in chess for a beginner?
- Play 1.e4, then develop your knights and bishops, castle, and avoid moving the same piece twice without a reason. 1.e4 opens lines for your queen and bishop, and our Maia3 data shows it is the move a model of players at 600 to 1000 picks most often, so you will meet familiar positions.
- Is 1.e4 or 1.d4 better?
- Stockfish 18 gives both +0.32. The difference is style: 1.e4 tends toward open, tactical games and 1.d4 toward slower, structural ones. Maia3 predicts 1.e4 is the more common choice at every band from 600 to 2400, but the gap narrows from 62.9% against 21.5% at 600 to 46.5% against 33.9% at 2400.
- What are the worst first moves in chess?
- By our Stockfish 18 depth 24 scores, 1.g4 is the worst at -1.26, then 1.f3 at -0.81, 1.Na3 at -0.66, 1.Nh3 at -0.62 and 1.h4 at -0.49. 1.f3 is also the first move of Fool's Mate, the shortest possible checkmate at 4 plies.
- Is 1.c4 or 1.Nf3 a good first move?
- Both are sound. Stockfish scores 1.Nf3 at +0.30 and 1.c4 at +0.17, and a model of players at 2400 plays them 8.2% and 6.0% of the time. They keep more options open than 1.e4 or 1.d4, which suits players who like flexibility, but they ask you to learn several setups.
- Does the first move decide the game?
- No. The best first move scores about a third of a pawn for White, and Black's reply changes that more than the choice among the top three. After 1.e4, for example, ...e5 scores +0.25 but ...d5 scores +0.71. We have no win rates by first move and rating, so we cannot say which first move wins most often at your level.
Try your first move against a model at your rating
Chessdrive runs Maia3 5m on-device as your opponent at any rating you set from 600 to 2600, so you can play 1.e4 or 1.d4 and meet the replies a player at your level tends to choose. A Stockfish review after the game shows where it turned. On iOS and Android, $5.99/month or $39.99/year on the US storefront with a free trial. Prices vary by region.
Keep reading
Updated