Stockfish and Maia3, measured
Chess openings by rating
Most opening pages tell you the best move. This one tells you how often a player at your rating actually finds it. Every legal move in 28 positions was scored by Stockfish at depth 20, then run through Maia3 at 7 rating bands to get the human answer.
Ordered by how hard the opening is to play well at 1000, hardest first. The percentage is how often Maia3 says a player at that rating plays the move Stockfish ranks first.
- Queen's Gambit Accepted1.d4 d5 2.c4 dxc4 3.Nf3Zero percent at 1000 and only 10 percent at 2200. This is a position to learn, not to work out at the board.1000: 11%2200: 30%+19 pts
- French Defence1.e4 e6 2.d4 d5 3.Nc3The lowest findability of any e4 opening in the run, under 20 percent at both ends of the rating range.1000: 19%2200: 25%+6 pts
- London System1.d4 d5 2.Bf43 percent at 1000, 14 percent at 2200. The London is easy to play and hard to answer well.1000: 21%2200: 23%+2 pts
- Scandinavian1.e4 d5 2.exd5 Qxd5 3.Nc318 percent at 1000 against 53 percent at 2200, one of the sharpest learning curves in the dataset.1000: 21%2200: 56%+35 pts
- King's Gambit1.e4 e5 2.f4Accepting the gambit is found about 40 percent of the time at every rating. The follow-up sits at 8 to 12 percent.1000: 26%2200: 26%+1 pts
- Scotch Game1.e4 e5 2.Nf3 Nc6 3.d4The widest opening gap in the whole run: 49 percent at 1000 against 96 percent at 2200 on a single capture.1000: 31%2200: 66%+35 pts
- Italian Game1.e4 e5 2.Nf3 Nc6 3.Bc4Findability barely moves from 1000 to 2200 because there is almost nothing to find. Both leading moves are correct.1000: 32%2200: 37%+5 pts
- Queen's Gambit Declined1.d4 d5 2.c4 e6 3.Nc3A smooth climb, 33 to 59 percent early and 44 to 70 percent late, with no rating band going backwards.1000: 38%2200: 64%+26 pts
- Sicilian Defence, Open1.e4 c5 2.Nf3 d6 3.d4 cxd4 4.Nxd447 percent at 1000 rising to 94 percent at 2200 on the early position, and 30 to 63 percent two plies later.1000: 39%2200: 78%+40 pts
- Vienna Game1.e4 e5 2.Nc3Flat and high across the range, 34 to 39 percent early and 54 to 59 percent late. Rating buys very little.1000: 44%2200: 49%+5 pts
- King's Indian Defence1.d4 Nf6 2.c4 g6 3.Nc3 Bg7Early findability runs 44 to 80 percent, but the position two plies deeper reverses it: 46 percent at 1000, 24 percent at 2200.1000: 45%2200: 52%+7 pts
- Caro-Kann1.e4 c6 2.d4 d5 3.Nc3Early findability climbs from 47 to 88 percent. Late findability falls from 45 percent at 1000 to 29 percent at 2200.1000: 46%2200: 59%+13 pts
- Englund Gambit1.d4 e553 percent at 1000 rising to 72 percent at 2200, and 63 to 82 percent on the follow-up. The trap mostly does not work.1000: 58%2200: 77%+19 pts
- Ruy Lopez1.e4 e5 2.Nf3 Nc6 3.Bb5Findability falls as rating rises, 28 percent at 1000 down to 15 percent at 2200. That is a repertoire choice, not a mistake.1000: 63%2200: 57%-5 pts
Your openings, your rating
Chessdrive runs Stockfish and Maia3 on your phone and applies exactly this to the games you play, so a missed move you were never going to find stops being called a mistake.
Method: Stockfish 18 by the Stockfish developers (see AUTHORS file) at depth 20 with MultiPV set to each position's exact legal move count, and Maia3, chessformer, human move prediction, 5m (bundled with maia3-js, UofTCSSLab/Maia3-5M, about 5M parameters). Run on 2026-08-29. Findability is the probability Maia3 assigns to Stockfish's top move at a given rating, not a win rate and not a claim that the other moves lose. The longer write-up is here.