Game Review benchmark
We reviewed 662 public games again, the way a review runs in your browser, and compared our accuracy with Chess.com's and Lichess's for each of the 1,324 players' games.
On this page
The results
On average our accuracy reads 1.2 points above Chess.com’s and 0.4 points below Lichess’s. One player’s game still reads a few points apart: 6.6 points on average against Chess.com, 4.1 against Lichess.
| Compared with | Players' games | Ours − theirs | Average gap | Correlation |
|---|---|---|---|---|
| Chess.com | 880 | +1.2 | 6.6 | 0.79 |
| Lichess | 444 | −0.4 | 4.1 | 0.93 |
Every player's game
One dot per player’s game. On the dashed line we gave the same accuracy as the other site; above it, ours read higher. Hover over a dot, or tap it, to see its numbers. Lichess gives accuracy as a whole number, so its dots stand in columns.
By time control and rating
| Against Chess.com | Players' games | Ours − theirs | Average gap | Correlation |
|---|---|---|---|---|
| All | 880 | +1.2 | 6.6 | 0.79 |
| Bullet | 48 | +2.4 | 6.4 | 0.85 |
| Blitz | 348 | +0.5 | 6.3 | 0.78 |
| Rapid | 472 | +1.6 | 6.8 | 0.78 |
| Daily | 12 | +0.6 | 4.4 | 0.76 |
| Rated under 1200 | 399 | +1.9 | 7.0 | 0.80 |
| Rated 1200 to 1599 | 133 | +0.9 | 5.8 | 0.71 |
| Rated 1600 to 1999 | 196 | +0.9 | 6.6 | 0.65 |
| Rated 2000 and up | 152 | −0.2 | 6.0 | 0.67 |
| Against Lichess | Players' games | Ours − theirs | Average gap | Correlation |
|---|---|---|---|---|
| All | 444 | −0.4 | 4.1 | 0.93 |
| Bullet | 276 | −0.4 | 4.4 | 0.93 |
| Blitz | 126 | −0.7 | 3.4 | 0.94 |
| Rapid | 34 | +0.5 | 3.8 | 0.94 |
| Classical | 8 | −1.5 | 1.8 | 1.00 |
| Rated under 1200 | 36 | −0.5 | 4.9 | 0.87 |
| Rated 1200 to 1599 | 106 | −0.1 | 3.8 | 0.94 |
| Rated 1600 to 1999 | 145 | +0.1 | 3.7 | 0.95 |
| Rated 2000 and up | 157 | −1.1 | 4.4 | 0.93 |
Ratings are each site’s own, and Lichess ratings generally run higher than Chess.com’s. A group of fewer than 30 players’ games is too small to read much into.
Why our accuracy changed
Our accuracy follows the method Lichess publishes: a move’s accuracy falls with the winning chances it gives away, and a game’s accuracy mixes a weighted average of its moves with one that its worst moves pull down. Until November 2026 we used Lichess’s settings unchanged, and on this benchmark they read 6.5 points above Chess.com: a game reviewed on both sites looked better here.
So we made the settings stricter. A move’s accuracy now falls faster with each point of winning chance given away (0.0625 per point instead of 0.0435), and a game’s accuracy takes 75% of the weighted average instead of 50%. We chose them on 70% of the Chess.com games, each 200-point rating band counting the same, and checked them on the other 30%: there ours read +1.6 against Chess.com, where Lichess’s settings read +6.6.
| Settings | Chess.com: ours − theirs | Chess.com: average gap | Lichess: ours − theirs | Lichess: average gap |
|---|---|---|---|---|
| Lichess's, until November 2026 | +6.5 | 9.3 | +4.9 | 5.2 |
| Ours, now | +1.2 | 6.6 | −0.4 | 4.1 |
Reviews saved before the change keep their old numbers in Insights until you open the game again.
Mistakes found
For each player in its games, Lichess publishes how many inaccuracies, mistakes and blunders it found. Counted Lichess’s way, our engine found 2.00 blunders per player per game where Lichess found 2.72. Part of that gap is the counting: Lichess also judges moves that throw away a forced mate, or allow one, by rules of its own. Our review grades with its own thresholds, which mark more moves as mistakes than Lichess’s do.
| Counted by | Inaccuracies | Mistakes | Blunders |
|---|---|---|---|
| Lichess | 3.22 | 1.54 | 2.72 |
| Our engine, Lichess's thresholds | 3.57 | 1.39 | 2.00 |
| Our review's grades | 4.25 | 3.32 | 2.41 |
Deeper searches
A Fast review searches every position to depth 14 on a laptop and to a little less on a phone; a Deep review then searches the moments that decided the game deeper. The same settings read alike at a phone’s depth and a laptop’s:
| Depth | Games | Chess.com: ours − theirs | Chess.com: average gap | Lichess: ours − theirs | Lichess: average gap |
|---|---|---|---|---|---|
| 12, a phone's Fast review | 662 | +1.5 | 6.6 | −0.2 | 4.3 |
| 14, a laptop's Fast review | 662 | +1.2 | 6.6 | −0.4 | 4.1 |
On 220 of the games we also searched every position to depth 18, deeper than any review’s first pass. With the depth the only change:
| Depth | Chess.com: ours − theirs | Chess.com: average gap | Lichess: ours − theirs | Lichess: average gap | Blunders found |
|---|---|---|---|---|---|
| 14 | +1.7 | 6.1 | −0.8 | 3.9 | 2.05 |
| 18 | +0.9 | 5.7 | −1.5 | 3.9 | 2.13 |
Game rating
The report gives each player a game rating, as Chess.com does: the game’s accuracy placed on the rating scale. One game says little about a player’s strength, though. On the Chess.com games, one game’s accuracy correlates only 0.37 to 0.45 with the player’s rating, depending on the time control.
So the game rating lays the accuracy scale on the rating scale, about 62 rating points per accuracy point, fitted on 858 Chess.com players’ games with each 200-point rating band counting the same. Players’ real ratings sat a median 525 points from their game rating. It reads one game, not a player’s strength, and the report shows it only for a side with 10 graded moves or more.
| Players rated | Players' games | Median game rating |
|---|---|---|
| 200 to 399 | 81 | 818 |
| 400 to 599 | 113 | 1,044 |
| 600 to 799 | 96 | 1,161 |
| 800 to 999 | 74 | 1,425 |
| 1,000 to 1,199 | 16 | 1,485 |
| 1,200 to 1,399 | 81 | 1,751 |
| 1,400 to 1,599 | 52 | 1,784 |
| 1,600 to 1,799 | 118 | 1,767 |
| 1,800 to 1,999 | 77 | 1,889 |
| 2,200 to 2,399 | 37 | 1,956 |
| 2,400 to 2,599 | 37 | 1,881 |
| 2,600 to 2,799 | 40 | 2,045 |
| 2,800 to 2,999 | 36 | 2,191 |
How we measured
- The games. 440 Chess.com games whose accuracy Chess.com published, which it does once anyone opens Game Review on a game, from players rated 100 to 3,056, found through players’ opponents, national player lists and titled players. And 222 Lichess games its server analysed, from players in recent arenas, rated 719 to 2,671. Standard chess only, at most three games per player.
- Our review. The engine build your browser runs, every position searched to depth 14 for its two best moves, then the same opening book, grading and accuracy code as the review screen. It ran on a computer with one search thread per game, where your browser uses several.
- What is compared. Each player’s accuracy in each game. Chess.com does not publish how it works out accuracy, so that comparison measures the whole result. Lichess publishes its method, which ours follows with stricter settings.
- When. The games were reviewed on 6 November 2026. We run the benchmark again whenever the engine, its depths or the grading change.
- Privacy. These are other players’ public games, so this page shows only totals and one anonymous dot per player’s game: no names, no links to the games.
Compare a game yourself
Review a game you have also reviewed on Chess.com or Lichess, and set the numbers side by side.