What the number actually measures
A win rate is the record of the games that reached one specific position, played by people in one rating range. It is a statement about those players, not about the position. An opening that scores 55% at 800 and 47% at 2000 has not got worse — it has stopped catching people out.
That distinction is the whole reason we publish the numbers by rating band rather than as a single figure. The band is the context that makes the number mean something.
Sample size is doing more work than you think
These figures are measured at the end of a twelve-move main line, and very few games get that far down any particular path. In our own dataset, individual cells range from a couple of dozen games to several thousand. At the low end, one result moves the percentage by several points, and a difference between two openings that looks decisive is noise.
The table below shows White's win rate for five openings across every rating band. Cells with fewer than 100 games are left blank rather than filled with a number that cannot support the weight — and how many blanks there are is itself the point.
| Opening | 400-600 | 600-800 | 800-1000 | 1000-1200 | 1200-1400 | 1400-1600 | 1600-1800 | 1800-2000 | 2000-2200 | 2200-2400 |
|---|---|---|---|---|---|---|---|---|---|---|
| Italian Game (White) | 46.3% n=108 | 46.6% n=779 | 52.8% n=316 | 59.0% n=122 | 49.0% n=2,839 | 49.0% n=5,279 | 50.3% n=664 | 48.0% n=854 | 45.4% n=864 | 46.5% n=630 |
| Scandinavian Defense (White) | — | — | 51.5% n=355 | 51.9% n=131 | 48.7% n=2,975 | 48.9% n=5,392 | 50.8% n=709 | 53.1% n=838 | 54.4% n=796 | 50.4% n=567 |
| Caro-Kann Defense (White) | — | 53.8% n=173 | — | — | 50.1% n=785 | 49.9% n=1,354 | 50.5% n=184 | 51.9% n=297 | 47.0% n=468 | 50.4% n=421 |
| Scotch Game (White) | 53.5% n=228 | — | 42.1% n=1,078 | 43.8% n=397 | 43.0% n=7,209 | 43.8% n=9,290 | 45.0% n=899 | 41.2% n=1,184 | 44.1% n=1,596 | 46.9% n=1,532 |
| King's Gambit (White) | — | 44.9% n=107 | — | — | 51.0% n=155 | 40.6% n=598 | 43.9% n=180 | 38.7% n=271 | 38.6% n=197 | — |
Three ways these numbers get misread
Picking the highest number. The best-scoring line in a rating band is often the sharpest one, which means it scores well because the defenders went wrong, not because it is sound. Play it against someone who has seen it and the number does not travel with you.
Ignoring who plays what. Openings are not randomly assigned. A line played mostly by well-prepared enthusiasts will show a good score that belongs to the players, not the moves.
Reading a draw rate as a boring opening. High draw rates at strong levels usually mean the position is well understood, not that it is lifeless — and at club level the same opening may be decisive nearly every game.
What they are good for
Comparing the same opening across bands, which tells you whether a line depends on your opponent's ignorance. Checking whether an opening you are considering actually appears at your level. And seeing which replies people really play — the most common continuation at 1200 is frequently not the one the theory books treat as the main line, and preparing for the move you will actually face is worth more than preparing for the correct one.
All of this data is available as an open dataset if you would rather check it yourself.
Frequently Asked Questions
What is a good win rate for a chess opening?
For White, anything near 50% is normal and healthy — White's small first-move advantage is worth a couple of percent, not a decisive edge. Be suspicious of anything far above that: it usually indicates a sharp line where the defence is hard to find, a small sample, or both, rather than an opening that is simply better.
Why do opening win rates differ so much between rating levels?
Because they measure the players, not the position. Lines that set an early problem score well where the problem goes unsolved and level off where it does not. That gap is useful information: an opening whose score falls away as ratings rise is one whose value is surprise rather than soundness.
Where does this data come from?
It is aggregated from Lichess game data through the same opening heatmaps that drive SpeakChess sparring bots, then published per rating band with the sample size attached. The full dataset is free to download under CC BY 4.0.