Lichess has more than six million puzzles, and most advice about them is a list of tips. We took a different route and read the whole database. The puzzle file is published under CC0, so we ran one pass over every row: 6,193,530 puzzles, each with a rating, a theme list, a solution and a play count. This page reports what that pass shows about what the trainer actually serves you as your rating climbs, then turns it into a plan.
One caution first, and it applies to every number below. These are puzzle ratings, the Glicko2 rating each puzzle earns from the people who attempt it, not player ratings. A puzzle rated 1600 is roughly one that a solver with a 1600 puzzle rating gets right about half the time. "Solver moves" means the moves you make; the first move of every puzzle is the opponent's.
What Lichess does well
It is worth saying plainly what Lichess gets right, because the rest of this page is about the one thing it leaves to you.
- It is built from real games. The database page says the puzzles come from 600 million analysed Lichess games, re-analysed with Stockfish NNUE and tagged automatically. The positions are not composed studies. They are moments that happened over a board.
- It is huge. 6,193,530 puzzles, with ratings from 399 to 3330. You will not run out.
- The data is open. The exports are CC0, which is why a page like this one is possible at all. Anyone can check our counts against the file.
- It is free. Lichess's about page calls the site 100% free, with no ads, and we found no puzzle cap. The contrast is real: as of its 4 September 2026 support article, Chess.com gives free members 3 rated puzzles a day.
- The theme system is deep. As of November 2026 the themes page has 11 sections and about 88 entries, including an origin filter for master games, super GM games and ordinary player games.
There are also several modes beyond the plain trainer. Puzzle Storm and Puzzle Racer both run alongside it, as of November 2026. We have not sourced the rules of either mode in detail, so we leave them there. For the one puzzle a day format, see our daily chess puzzle guide.
The shape of the whole database
Start with the overall distribution of puzzle ratings. The median puzzle is rated 1411 and the mean is 1467.9. Ten percent of puzzles are rated 800 or below and ten percent are rated 2241 or above. The middle half sits between 1038 and 1870.
Solutions are short. Half of all puzzles, 50.58%, need exactly 2 moves from you. Another 26.29% need 3, 14.88% need 1, and 5.88% need 4. Beyond that the numbers fall away quickly: 1.59% need 5 and 0.5% need 6. A Lichess puzzle is almost never a long forced march. It is a two or three move idea you either see or do not.
50.2% of puzzles carry the endgame tag, 50.16% are tagged short, 45.04% middlegame and 38.44% crushing. Among actual tactics, mate in 1 is on 14.76% of puzzles, mate in 2 on 13.24%, forks on 12.84%, sacrifices on 7.54% and pins on 6.0%. A puzzle can carry several themes, so these shares overlap and do not add to 100.
Each theme also has a typical difficulty. Back rank mate has a median puzzle rating of just 784, mate in 1 is at 908, and mate in 2 at 1083. At the other end, quiet move puzzles have a median rating of 2123, very long puzzles 2110 and defensive moves 2041. Forks sit in the middle at 1311 and pins at 1637. Difficulty is not only about how many moves you must find. It is about what kind of move.
What you meet at each rating band
This is the table the rest of the page builds on. Each row is a band of puzzle ratings. "Share of puzzles" is how much of the database falls in that band. "Share of plays" is how many of the 10,975,127,289 recorded attempts were made on puzzles in that band. The theme columns are the percentage of puzzles in the band that carry that theme.
| Puzzle rating | Share of puzzles | Share of plays | Median solver moves | Mate in 1 | Mate in 2 | Fork | Sacrifice | Quiet move |
|---|---|---|---|---|---|---|---|---|
| Under 1000 | 22.85% | 6.04% | 2 | 41.2% | 24.5% | 13.5% | 3.1% | 0.7% |
| 1000 to 1199 | 14.4% | 10.22% | 2 | 19.2% | 20.3% | 16.5% | 5.1% | 0.8% |
| 1200 to 1399 | 11.95% | 18.25% | 2 | 11.5% | 15.7% | 14.5% | 7.0% | 1.4% |
| 1400 to 1599 | 11.72% | 24.97% | 2 | 6.9% | 11.1% | 13.8% | 9.3% | 2.5% |
| 1600 to 1799 | 10.79% | 16.93% | 2 | 2.9% | 7.9% | 12.5% | 10.3% | 3.9% |
| 1800 to 1999 | 9.26% | 11.7% | 3 | 0.8% | 5.2% | 11.2% | 10.6% | 5.8% |
| 2000 to 2199 | 7.67% | 6.93% | 3 | 0.1% | 2.1% | 9.5% | 11.0% | 8.5% |
| 2200 to 2499 | 7.57% | 3.63% | 3 | 0.0% | 0.5% | 8.5% | 10.8% | 13.0% |
| 2500 and up | 3.78% | 1.33% | 3 | 0.0% | 0.0% | 7.1% | 11.6% | 22.4% |
Three things stand out.
Mate in 1 disappears. It is on 41.2% of puzzles rated under 1000, then 19.2%, 11.5%, 6.9% and 2.9% as you climb, and by 2200 and up it is 0.0%. The same goes for back rank mate: 10.7% of puzzles under 1000, 3.7% at 1000 to 1199, 0.4% at 1600 to 1799, and 0.0% at 2200 and up. Mate in 2 follows the same slope, from 24.5% to 0.0% at the top band.
Quiet moves take over. A quiet move is a solution that does not capture, check or make an obvious threat, and it is what players at a higher level fail to find. It sits on 0.7% of puzzles under 1000 and 22.4% at 2500 and up. Defensive move puzzles go from 0.6% to 23.8% over the same range, and very long puzzles from 0.8% to 40.6%. The high end of Lichess is a different kind of problem, not a harder version of the low end.
Solutions get one move longer. The median is 2 solver moves in every band below 1800 and 3 from 1800 up. The mean tells the same story more smoothly: 1.67 moves under 1000, 2.51 at 1600 to 1799, and 3.55 at 2500 and up.
Where the plays actually are
Share of puzzles and share of plays are very different columns. The three bands from 1200 to 1799 hold 34.46% of the puzzles (11.95 + 11.72 + 10.79). They absorb 60.15% of all plays (18.25 + 24.97 + 16.93). The single busiest band, 1400 to 1599, takes 24.97% of plays with 11.72% of the puzzles.
The bottom band shows the opposite. Puzzles rated under 1000 make up 22.85% of the database and get 6.04% of the plays. There is a large pool of easy, mostly mate in one or two material that comparatively few attempts are spent on. The dump does not say who played what, so we can only guess at why. The natural reading is that most people spend their time on puzzles near their own puzzle rating, and that a lot of the active population sits in the 1200 to 1799 range.
Plays are also concentrated in a few puzzles. The top 1% of puzzles take 17.87% of all plays, and the median puzzle has been played 404 times.
How reliable is a puzzle rating?
Each puzzle carries a rating deviation, the Glicko2 measure of how unsure the system is about its rating. In the current file, 12.74% of puzzles still have a rating deviation over 100. That is about one in eight. Those puzzles have not been attempted enough for the number to settle, and their rating may drift.
If a puzzle feels far easier or harder than its rating suggests, it may be one of these young ones. A puzzle rating is a good guide across hundreds of attempts and a poor one across ten.
The daily puzzle is harder than the typical one
The dataset has a daily puzzle date column, and 1,812 puzzles in the dump carry one. Their median puzzle rating is 1972, with the 10th percentile at 1795 and the 90th at 2121. Compare that with the database as a whole: the 75th percentile is 1870. The median daily puzzle is therefore harder than three quarters of everything in the trainer.
So if the puzzle of the day usually beats you, that is not a sign that you are falling behind. The pool it is drawn from is narrow and hard. We cover how a daily habit compares with a retention system in our guide to the daily chess puzzle habit. We have not sourced exactly how Lichess picks the daily puzzle, so we make no claim about that process.
What the theme mix means for you
The table works as a map of the kind of problem you will meet. Here is how we would read it at each level. These are our reading of the data, not Lichess advice.
Under 1000. Four in ten puzzles are a mate in one, a quarter are a mate in two, and 13.5% are forks. Back rank mates are 10.7%. The skill that pays here is a habit: before anything else, list every check you have and every capture. Most of these puzzles are solved by a forcing move you were allowed to see. Playing them quickly is fine. Playing them carelessly is the usual failure. For a linked, theme by theme route through this band, see our chess puzzles for beginners ladder.
1000 to 1399. The mate in one share falls from 19.2% to 11.5%, while forks peak at 16.5% in the 1000 to 1199 band. Pins reach 5.2% and then 7.0%, discovered attacks 6.1% and 6.5%, and deflection 4.9% and 5.6%. This is the band where tactics stop being single ideas and start to need a two move plan. If you only practise forks, you will meet the others in your games and not recognise them.
1400 to 1799. This is the busiest region, and the mix is the most balanced. Sacrifices go from 9.3% to 10.3%, attraction sits near 5.9%, and quiet moves rise from 2.5% to 3.9%. Mate in 1 is now 6.9% and then 2.9%, so checkmate patterns are no longer the main event. Your gains here come from calculating a sacrifice to its end and from seeing the quiet move after the obvious check does not work.
1800 to 2199. The median solution is now 3 solver moves. Quiet moves go from 5.8% to 8.5%, defensive moves from 9.4% to 13.2%, and very long puzzles from 11.6% to 16.9%. Many of these are about not losing a good position or finding the one accurate waiting move.
2200 and up. Mate in 1 is 0.0%. Quiet moves are 13.0% and 22.4%, defensive moves 17.1% and 23.8%, and very long puzzles 26.0% and 40.6%. Zero in on calculation depth. A rating here is mostly a measure of how well you see non obvious moves several plies ahead.
How to use the theme list on purpose
The themes page is the most useful tool in the trainer, and also the easiest to misuse. Here is the plan we would follow.
- Start with the healthy mix. The themes page puts a recommended mix at the top for a reason: you do not know in advance which motif a game will hand you. A mix trains recognition. A single theme trains the response once you already know the theme.
- Know what a theme filter gives away. If you pick the fork theme, you know before you start that a fork exists. In a real game nothing tells you. Theme practice is good for learning what a pattern looks like. Mixed practice is what tests whether you notice it. Alternate the two.
- Use your results to choose themes. A forum reply points to the Puzzle dashboard on your profile for 30 day progress and strengths and weaknesses by theme. Find the themes where you fail most often, then practise those, not the ones you enjoy.
- Match the theme to your band. From the table: a 1100 player gains little from quiet move puzzles, which are 0.8% of puzzles at that level, and a 2000 player gains little from mate in one. Pick themes common in the band just above yours.
- Link puzzles to your own games. The puzzles that transfer best are the ones that look like your games. See our Lichess analysis walkthrough for finding your own mistakes, and our piece on puzzles that transfer to games for why puzzle rating and game rating often disagree.
If you want the underlying tactics rather than the trainer, our chess tactics guide for improvers covers each motif with examples.
Where Lichess stops
Everything above is about selection: what the trainer puts in front of you. The trainer is very good at that. What it does not do, as far as we could confirm, is decide that the specific pattern you just missed should come back to you in a few days.
That matters because the database is so large. With over six million puzzles, meeting the same position again by accident is unlikely. You can fail a back rank pattern on Monday and, in the ordinary flow of new puzzles, not see anything like it for weeks. The trainer serves a stream of new positions. Remembering the ones you missed is your job.
There are manual tools. A forum reply describes replaying failed puzzles from the dashboard, and we could not confirm the current label, so check your own profile. That is a list you have to remember to open, on a timetable you set yourself. If you use it, a simple rule helps: when you fail a puzzle, write down the theme and the key move in a sentence, and redo it after a day, then after a week.
Where Chessdrive fits
Chessdrive does the part that a puzzle stream does not. It builds puzzles from the mistakes in your own games, and it brings a missed puzzle back on a spaced schedule instead of leaving it to chance. It does not replace Lichess for volume or for the range of themes. We would use Lichess to find what you are weak on, and a scheduler to keep what you have learned.
Method and limits
The figures in this article come from one pass over the Lichess puzzle CSV, the CC0 file at database.lichess.org, in the copy we downloaded on 5 November 2026. We read 6,193,530 rows, the same count the database page lists. Solver moves are the moves in the solution divided by two, because the first move is the opponent's. All ratings are puzzle ratings. Theme shares overlap, because a puzzle can have many tags, and the band shares are rounded to two decimals as in the source file. The data does not show who made the attempts, or whether practising a theme improves game results.
Frequently asked
- Are Lichess puzzles free?
- Yes. Lichess describes itself as 100% free with no ads, and the whole puzzle database is published under the Creative Commons CC0 license, so anyone can download and reuse it. We found no daily puzzle cap. For contrast, Chess.com's support article, as of 4 September 2026, gives free members 3 rated puzzles a day.
- Where do Lichess puzzles come from?
- From real games. The database page says the puzzles were built from 600 million analysed Lichess games, with positions re-analysed by Stockfish NNUE and then tagged automatically, and that player votes refine tags and popularity. That is why many puzzles, especially at lower ratings, look like positions that arise in ordinary play.
- How does the Lichess puzzle rating work?
- Each attempt counts as a Glicko2 rated game between you and the puzzle, so both you and every puzzle carry a rating and a rating deviation. Beat a hard puzzle and you gain more than you do for beating an easy one. We did not verify the starting rating or the exact gain and loss sizes, so we do not quote them. The point to remember is that a puzzle rating measures puzzle solving, not your playing strength.
- What puzzle themes does Lichess have?
- The themes page groups puzzles into 11 sections, about 88 entries in our check: a recommended healthy mix, phases, openings, tactical motifs, advanced themes, mates, mate patterns, special moves, goals, lengths and origin. The origin filter separates master games, master versus master, super GM games and ordinary player games.
- Why is the Lichess daily puzzle so hard?
- Because the ones that were picked as daily puzzles skew hard. Of the 1,812 puzzles in the dump that carry a daily puzzle date, the median puzzle rating is 1972, against 1411 for the whole database. Only the top quarter of the database sits above 1870. If today's puzzle feels out of reach, that is normal and says little about you.
- Can I replay puzzles I failed on Lichess?
- A Lichess forum reply says you can replay failed puzzles from the Puzzle dashboard on your profile page, alongside 30 day progress and strengths and weaknesses by theme. We could not confirm the current button label, so look for it on your own dashboard. Replaying is manual: you choose when to go back, and nothing schedules it for you.
- Do I need an account to solve Lichess puzzles?
- You can solve them without one, but a third party description of the puzzle feature says progress is not saved without an account. Sign in for a rating and history.
Turn the puzzles you miss into puzzles you keep
Chessdrive builds puzzles from your own games and brings back the ones you got wrong on a spaced schedule, so a miss is not lost in a stream of new positions. Pair it with Lichess for volume. Free to start, on iOS and Android.
Keep reading
Updated