Play chess online free and the honest answer is Chess.com or Lichess, both loading the moment you open the tab, neither one asking for a card. Lichess is entirely free, entirely ad free, and never puts a paywall anywhere in the experience. You can even play a person or a computer without creating an account. Chess.com also plays free in the browser and allows a handful of games without an account too, with its paid tiers gating a daily game cap and a few extra features rather than access itself. Chessdrive, the app behind this article, is not something that opens in a tab: it's a downloaded iOS and Android app, an AI chess coach that runs on your device. So when you search for this, the honest answer starts with Chess.com and Lichess, and what follows is what each one gives you and how to pick.
- Stockfish best
- Qd6-6.76
- Players at 1400 play
- h644%
Where to actually play chess online for free
There are two real options, and both work today, right now, with no account required.
- Lichess.org: Open source, ad free, donation supported. No account is needed to play a person or a computer. You can run full Stockfish analysis server-side on any game you want (on demand, within a daily cap), and there is no paid tier anywhere on the site at all. It stays the cleanest, most honest free option available.
- Chess.com: Also plays free in the browser, allows a limited number of games without an account, and carries by far the larger player base and social features of the two. Its free tier is real, but the daily game limit and some analysis features sit behind a subscription, which is the actual tradeoff against Lichess rather than any hidden cost to entry.
Beyond those two, a few other names turn up in search results, and each deserves an honest note rather than a blanket recommendation. Generic gaming portals like CrazyGames or Poki will drop you into an instant, no-login chess game, but you're playing against a very basic bot with no real teaching value and no rating system worth trusting. Sites that market themselves as brain training platforms, the kind that lead with a "play chess online" headline but funnel you toward a sign-up wall a click or two in, are cognitive assessment products first and chess sites second. Know that before you invest ten minutes filling out a profile. If a result looks too polished with too little actual content behind the landing page, that's usually the tell.
How to play chess
The board is 8x8 squares, and each side starts with 16 pieces: 1 king, 1 queen, 2 rooks, 2 bishops, 2 knights, 8 pawns. The goal is to checkmate the opponent's king, meaning the king is under attack with no way to escape, block, or capture the attacker. White always moves first.
| Piece | Count | How it moves |
|---|---|---|
| King | 1 | Any direction, one square only |
| Queen | 1 | Any direction, straight or diagonal, any number of squares |
| Rook | 2 | In a straight line, horizontal or vertical, any number of squares |
| Bishop | 2 | Diagonally, any number of squares |
| Knight | 2 | In an L shape: two squares one direction, one square at a right angle |
| Pawn | 8 | One square forward, two on its first move, captures diagonally |
There are three special moves. Castling is a move the king and one rook make together, only when there are no pieces between them and neither has moved yet. You also cannot castle out of check, through a square an enemy piece attacks, or into check. The king slides two squares and the rook jumps over it to the other side. En passant lets a pawn capture a rival pawn that just advanced two squares past it on its first move, as though it had only moved one square, but only in the single turn immediately after that advance. Pawn promotion happens when a pawn reaches the far end of the board and turns into a queen, rook, bishop, or knight. Almost everyone promotes to a queen, since it's the strongest piece.
A game ends one of three ways. Checkmate ends it outright, one side loses. Stalemate occurs when the player to move has no legal move at all and their king isn't under attack: the game is a draw. Agreed or forced draws also happen through threefold repetition, the same position occurring three times, or the fifty-move rule, fifty moves passing with no capture and no pawn moved.
A classic trap that still catches beginners today is Scholar's Mate: 1.e4 e5 2.Bc4 Nc6 3.Qh5 Nf6?? 4.Qxf7# ends the game in four moves. It only works if the defender misses the queen's early attack on f7. Any of ...Qe7, ...Qf6, or ...g6 shuts it down completely.
Once you know the rules, three principles are worth keeping in mind for your opening moves: control the center, meaning claim squares like e4, e5, d4, or d5 early; develop your pieces quickly, bringing a new piece into play each move rather than moving the same one repeatedly; and castle early to get your king to safety. None of these are hard rules, just practical habits. Traps like Scholar's Mate work precisely against players who violate them, bringing the queen out too early in this case.
Two player chess and playing offline
Playing across the same board is the simplest version of two player chess: with a physical set, you never need internet at all. For playing two player online, both Lichess and Chess.com let you invite a friend to a private game link you can play in real time, and you can use that link on Lichess without ever creating an account.
Playing offline is a slightly more nuanced question. The surest route is still a physical board: a standard Staunton set and a folding board are never affected by an app update or a server outage. Among apps, what varies is where the engine runs: some call a server for every single move and simply won't function without a connection. In Chessdrive, Maia3 and Stockfish both run on your device, so playing against the computer never sends move calculation to a server. We're not claiming every other feature, like account login or sync, works offline too, only that the move engine itself is on-device.
Why playing against a computer so often feels wrong
Search for the best chess AI and a lot of what comes back still leans on the 2017 AlphaZero result, where Google's model beat Stockfish 28 wins to 0 losses across 100 games, with 72 draws. That was true in 2017. Two things have changed since, and most search results still barely mention either.
First, raw strength. Stockfish 18 released on 31 January 2026, introducing the SFNNv10 neural network, new Threat Inputs features, and shared-memory handling for multi-threaded efficiency. That delivered an Elo gain of up to 46 points over Stockfish 17. Still free, still open source.
Second, on a completely different axis: human likeness. Built on an architecture the University of Toronto's Computational Social Science Lab calls Chessformer, Maia3 is an encoder-only transformer using squarewise tokens and Geometric Attention Bias. Its ICLR 2026 paper reports the 79 million-parameter version matching real human moves 57.1 percent of the time, and the 23 million-parameter version at 56.6 percent. Both beat the prior best model, a 355 million-parameter system called ALLIE that scored between 55.7 and 55.9 percent, using under a quarter of the parameters. Even the smallest 5 million-parameter version reaches 55.4 percent, matching the old best model with roughly 70 times fewer parameters. All three are free, open source, and public on Hugging Face and GitHub.
Stockfish 18 and Maia3 don't answer the same question. One answers what's the strongest move here; the other answers what would a real human at this rating actually play. That's exactly why so many depth-limited, easily adjusted computer opponents feel off: they either fold too early or throw in random moves no real person would ever make, because capping a search depth isn't the same thing as playing like a human. We cover why that happens, and how Maia tries to fix it, in our companion piece on Maia chess. Maia's first generation trained as nine separate models, but only three are actually playable as Lichess accounts: maia1, maia5, and maia9, all free. These bots play a bit stronger than their labels suggest: they skip the characteristic mistake roughly 20 percent of players at that rating would make in a given position, while still repeating other errors typical of that level. CSSLab runs no official Lichess account for Maia3. The only live Maia3 bot we could find on 28 August 2026 was a community-run account, @maia3-79m_2200, not an official one.
The practical test when picking a bot: does playing against it feel like playing a real person near your level, or does it either crush you or fold before you do? The data below shows the mechanics behind that feeling through a single worked example.
A move being the best doesn't mean it's findable
An evaluation score tells you a move is the best one. It doesn't tell you whether someone at your rating would actually find it on the board, and we can show exactly how far apart those two things can be using our own data.
We took two positions from the finish of Morphy's famous Opera Game against the Duke of Brunswick and Count Isouard, Paris 1858. We ran the on-device 5 million-parameter version of Maia3, matching every legal move against the score Stockfish 17 gives it at depth 20 using MultiPV, across seven separate self-rating bands: 1000, 1200, 1400, 1500, 1800, 2000, and 2200. What we're measuring is the probability Maia3 assigns to the move Stockfish rates as best, in other words, how likely a human at that rating actually is to play it. We call this findability. It isn't the same thing as Maia3's published 57.1 percent overall move-match accuracy. That figure measures predicting human moves across thousands of real games, while this measures one game's single sequence with the small 5 million-parameter model. A small sample, and we're saying so in the same sentence as the numbers.
Two positions from that same finish make the gap as clear as possible. The first is the final move, 17.Rd8#, mate in one. The second is three moves earlier, move 14 with Black to move, where Stockfish's best defense is an odd-looking queen move, Qd6.
| Rating | 17.Rd8#, mating move, findability | Qd6, three moves earlier, defense, findability |
|---|---|---|
| 1000 | 92.65% | 0.77% |
| 1200 | 96.10% | 0.70% |
| 1400 | 97.75% | 0.69% |
| 1500 | 98.21% | 0.69% |
| 1800 | 98.94% | 0.71% |
| 2000 | 99.11% | 0.73% |
| 2200 | 99.09% | 0.73% |
Even a 1000-rated player finds the mating move 92.65% of the time, climbing to 99.09% at 2200. Qd6, meanwhile, stays under 1% at every single band from 1000 to 2200, never rising above 0.77%. Rating barely moves the needle. What players at 1400 actually play most often in that position isn't Qd6 at all: it's h6, at 44.1%.
That's one finding from one game, one short sequence, but it carries the whole argument. In the same finish, one engine move gets found by ninety-three to ninety-nine players out of a hundred, regardless of rating, while another gets found by fewer than one in a hundred at every rating from 1000 to 2200. An evaluation score only tells you a move is best. It doesn't tell you whether someone at your level would actually find it on the board, and that gap is measurable.
The practical takeaway: when an analysis tool tells you a move was the best one, how instructive that sentence actually is depends heavily on the position. On a move nearly everyone finds anyway, like the mate, the engine's approval teaches you nothing new. You'd have played it regardless. On a move almost nobody finds, like Qd6, the engine's approval is accurate but describes something you shouldn't expect to reproduce in your own games. Without a number showing the difference between those two cases, they look identical.
Chessdrive, the party generating this measurement, has a stake here: the app runs Maia3 on your device as the opponent, and Stockfish separately as an analysis layer. Both the interface and the AI coach respond in whichever language you set.
Last verified: 28 August 2026, against the sources cited above and Chessdrive's own fresh engine output. Search volume, engine versions, and third-party sites' own features change without notice. Confirm current details on each platform's own page before publishing.
Frequently asked
- How do you play chess?
- Chess is played on an 8x8 board, each side controlling 16 pieces: 1 king, 1 queen, 2 rooks, 2 bishops, 2 knights, 8 pawns. The goal is to checkmate the opponent's king. White always moves first, then players alternate. Every piece has its own movement pattern, covered piece by piece in the rules section above. Castling, en passant, and pawn promotion are the only special moves.
- Where can I play chess online for free?
- Lichess.org and Chess.com both let you play in a browser with no account and no charge. Lichess is completely free with no ads and no paid tier anywhere on the site. Chess.com also plays free in the browser, with a generous but not unlimited daily game allowance before its paid tiers kick in. Both are legitimate, both are safe, and neither requires a download to start playing right now.
- What are the benefits of playing chess?
- The most commonly cited benefits are that chess exercises both analytical and creative thinking at once, and that memorizing lines and combinations strengthens short-term and long-term memory. Researchers have also noted correlations between learning chess young and academic performance in observational studies. Treat these as widely repeated observations rather than settled clinical fact, since the studies behind them are observational, not randomized proof of causation.
- How do you play two player chess?
- There is no rule difference from sitting across a physical board and alternating turns. If you want to play two player online, Lichess lets you send a friend a private game link with no account required, and Chess.com offers a similar invite feature.
- What is the best chess engine or AI?
- It depends on the question you're actually asking. For raw strength, the answer is Stockfish 18, released January 2026, free and open source. For playing like a human, the answer is different: the University of Toronto's Maia3 model matches real human moves 57.1 percent of the time, also free and open source. A lot of pages answering this question still lean on the 2017 AlphaZero versus Stockfish match, which is dated and no longer describes either project's current state.
- Can you play chess offline?
- The surest way is a physical board. Among apps, the difference comes down to where the engine runs: some call a server for every single move. In Chessdrive, Maia3 and Stockfish both run on your device, so move calculation itself never touches a server. We're not claiming every other feature, like account sign-in, works without a connection, only that the move engine does.
- How long does it take to learn chess?
- You can learn the rules in one sitting. How each piece moves and what checkmate means takes maybe thirty to forty minutes to grasp. Playing comfortably without giving away obvious material takes a few weeks of regular practice. A real improvement curve, where most of your moves start resembling what an engine near your rating would suggest, takes months or years of consistent play and review. Giving a shortcut number for that wouldn't be honest.
Looking for what comes after the free games against a computer?
Chessdrive runs Maia3 on your device with an adjustable rating from 600 to 2600, so it plays like a real opponent near your level, mistakes included, rather than something that either resigns early or plays inhumanly perfect chess. Stockfish sits alongside it as a separate review layer once the game is over. Available on iOS and Android.