Lumichess puzzles are drawn from the Lichess open puzzle database: 5,029,552 positions, each one taken from a real game where a player had a forcing continuation available. They are sorted into 15 rating bands running from 200 to 3199, so there is a workable difficulty level whether you are learning how a fork works or calculating six-move sequences.
Every mode is free, and you do not need an account to start solving.
A puzzle position is not a random arrangement of pieces. It is a moment from a real game where one side missed something concrete, and the solution is a forcing line — a capture, a check, or a threat the opponent cannot answer. Training on them builds two separate skills that are easy to confuse.
The first is pattern recognition: seeing that a position contains a pin or a loose piece before you have calculated anything. This is fast, close to automatic, and it improves mostly through volume. The second is calculation: working out the actual sequence, including the opponent's best replies, and confirming it works before you commit. This is slow and effortful, and it improves through difficulty rather than volume.
The modes below train these in different proportions, which is why using more than one is more useful than grinding a single mode.
Most puzzles below roughly 1800 resolve to a handful of recurring shapes. Recognising them by name makes them easier to spot over the board:
Rated training is the default. Puzzles are matched to your current rating, and your rating moves with your results, so difficulty tracks your level without you having to pick it. Use it when you want steady progress.
Puzzle Storm is timed. Solve as many as you can before the clock runs out, with easier puzzles early and harder ones as your run continues. It trains recognition speed rather than depth, and it is the mode most likely to expose patterns you know in theory but do not yet see instantly.
Woodpecker training takes a fixed set and has you solve the same puzzles repeatedly, faster each cycle. The point is not to solve them — you will know the answers by the third pass — but to move the patterns into recall so they surface without calculation.
Mistake training builds puzzles from positions in your own games where you went wrong. These are the most directly useful, because they target errors you have actually made rather than errors in general.
The daily puzzle is one position a day, the same one for everyone. It is a habit anchor more than a training method.
Puzzle rating uses Glicko-2, the same rating system Lichess uses for its puzzles. Each puzzle has a rating, and solving one rated above you gains more than solving one below you. Alongside your rating the system tracks a deviation value, which represents how confident it is in that number: new solvers move quickly because their deviation is high, and ratings settle as more puzzles are solved.
A puzzle rating is not a game rating and the two are not directly comparable. It measures how well you solve positions where a tactic is known to exist, which is an easier problem than finding one in a game where you do not know whether anything is there.
Two habits make a large difference. First, calculate the full line before you play the first move. Moving quickly and correcting after seeing the response trains guessing, not calculation. Second, when you get one wrong, work out what you missed before moving on — usually it is an opponent reply you did not consider rather than an idea you did not know.
If you want to know which motifs you specifically miss, a Profile Report analyses your recent games and reports where your tactical losses actually come from. To review a single game move by move, use Game Report.