Chess Bot to Play Against: Pick One by Your Rating

Ask which chess bot to play against and you'll get three different kinds of answers depending on where you look: a numbered difficulty level, a wall of a hundred named personalities, or a single dial that claims to already know your rating. None of those answers is wrong, but none of them actually tells you which specific bot fits a player at your own level today. That's the gap this page fills: a rating band matrix naming real, currently active bots on Lichess and Chess.com, a couple of smaller options worth knowing about, and then what actually separates a bot worth your time from one that just wastes it.

Part of the difficulty is that neither big site actually curates for this question. Lichess's own bots directory lists 397 active accounts with no filtering by strength or style at all, which is comprehensive but means the handful of genuinely human-like ones, the Maia family, LeelaRogue, the dala series, are effectively invisible unless you already know their names going in. Chess.com's own help center names only two or three bots with any rating attached at all, and never says which of its 100-plus options actually suits a given rating. Neither gap means the right bot doesn't exist on either site. It means nobody built the matrix first.

The rating band matrix: which bot, and where

Five bands cover most players, from a first game through strong club level. Every bot named below was active as of a mid September check; treat the exact numbers as a snapshot rather than a permanent fact, since rosters and ratings on products we don't control shift without notice.

Chessdrive is the odd one out in every row of that table: instead of a bot name, you set one continuous rating from 600 to 2600 and the app runs the real maia3-5m checkpoint on-device at that number, so it belongs in every band above rather than a single row of it. That's a narrower claim than either Lichess or Chess.com makes, and it's worth saying plainly before anything else: Lichess is free and unlimited with no paid tier anywhere on the site, and Chess.com's roster, over a hundred named bots, is bigger than anything Chessdrive or the smaller options above offer. If the widest possible free choice is what actually matters to you, pick one of those two first.

Worth knowing too: CSSLab's own site, maiachess.com, isn't just a mirror of the Lichess bots. It adds a puzzles mode, a Hand and Brain mode, and a Bot or Not Turing test style game built around the same models, signed in through your own Lichess account, with no pricing or paywall visible anywhere on it.

Why a bot's listed rating and its real difficulty rarely match

Chess.com's own community has argued this out in detail on its forums, and the consensus is blunt: deduct roughly 200 to 300 points from whatever number a bot displays before trusting it. Antonio's own page reads 1500; a Chess.com forum moderator's own estimate puts its real play closer to 1000. Isabel reads 1600, and at least one player reports beating it at their own roughly 850 rating. Wally reads 1800 and plays, by the same community consensus, closer to 1500. Xavier is the clearest example of why a single number can't describe a whole game: listed at 1900, it opens with roughly eight moves of genuine opening book knowledge, then hands the rest of the game to whatever move it judges objectively best, which plays like a considerably stronger opponent than its own listed number promises.

Lichess's own bot ratings move too, just for a different reason: the displayed number reflects whoever has played that bot recently rather than a fixed calibration, so BOT maia1's own page can read anywhere from the low 1400s to about 1600 depending on the format and time control you check. Neither site's displayed number is a fixed fact; treat it as a rough starting point and adjust from what actually happens over the board. For what a rating number is actually measuring in the first place, our own explainer on the chess rating system covers that separately.

A quick way to sanity check any bot before committing a whole session to it: play three or four moves you already know are natural for a player at that level, then watch what happens once the position leaves the opening. If a bot that reads 1800 suddenly starts finding only ever the objectively correct answer a few moves in, the way Xavier does, its displayed number is really describing two different opponents stitched together, not one consistent level.

None of this unpredictability is really a flaw. Every one of these products makes an already strong engine play worse on purpose, whether by capping Stockfish's own Skill Level and search depth or by authoring per-bot settings on Chess.com's Komodo engine, and that kind of capping is a fundamentally different mechanism from how a genuinely weaker human actually thinks through a position. We cover exactly what a difficulty setting changes under the hood, and why it produces bots that play well for a while and then blunder without warning, on our page about how to choose a computer opponent; it isn't repeated here.

What actually makes a bot worth playing

Three things separate a bot worth your time from one you'll quit on within a week.

• Human shaped mistakes, not random ones. A bot that plays five accurate moves and then hangs a piece for no visible reason doesn't feel like a weaker opponent, it feels broken. Maia was built specifically to avoid that failure by training on real human games at a target rating instead of capping a search engine, which is why its errors read as a real person's rather than noise. We cover the research behind that gap in full in why chess bots play five good moves and then hang a queen and in our own reference on Maia chess.
• Whether hints and takebacks are actually available. Chess.com ties both to a reward system rather than blocking them outright: finish a bot game with no help used and you earn three crowns, use one to three hints or undos and you earn two, use four or more and you earn one. The help is real, it's just gamified rather than free of any cost.
• Whether there's a review layer once the game ends. Chess.com's own Game Review, or Chessdrive's separate Stockfish pass after the game closes, is what turns a loss into something you actually learn from rather than just a result on the board.

Two smaller competitors worth naming here, honestly, rather than glossing over. Chessiverse markets what it calls PersonaPlay technology, over 1000 named personalities and a claimed 1 million plus calibration games, but names neither Maia nor Stockfish anywhere on its own page and cites no published research behind its human likeness claim, so there's no way to check how it actually works under the hood. Noctie.ai takes the opposite approach: an auto adaptive bot that claims to detect your playing strength automatically rather than offering a named roster at all, but it also discloses no underlying engine and publishes no concrete Elo numbers for its own bot, so you're trusting its word on both fronts.

When to actually move up a band

Winning against a bot isn't, on its own, a great signal to move up: a bot this weak still blunders on its own schedule, and beating one a few times in a row can just mean you got lucky with what it happened to hang. A better signal is whether you're finding moves your current bot's own rating band would still miss.

We tested this directly rather than guessing. In our own Maia3 run on the finish of the 1858 Opera Game, Morphy against the Duke of Brunswick and Count Isouard, we measured how likely Maia thinks a player at each rating is to find one specific move: Morphy's 14.Rd1, a quiet move that brings his second rook into the attack rather than a forced recapture. At a 1000 rated band, that move gets found 36.70% of the time. At 1200, 42.16%. At 1400, 51.71%. At 1500, 56.94%. At 1800, 70.91%. At 2000, 77.34%. At 2200, 81.11%. The curve climbs steadily rather than jumping at any one point, and that's the useful part. A 1000 rated player misses that move nearly two times in three; a 1400 rated player misses it about half the time; by 2000, fewer than one time in four. It is one position, not a benchmark, but the shape is the point: strength shows up as finding the quiet, strong move more often, not as winning more often against an opponent that hangs pieces. If you're finding moves like that and your bot keeps missing them, that's a cleaner signal to move up than a plain win streak.

Picking without hunting for a bot's name

If you already know which site you want to play on, the matrix above should get you to the right bot. If you'd rather skip picking a name entirely and just set a single rating that moves with you, that's the narrower thing Chessdrive is built to do: the real maia3-5m checkpoint on-device, one continuous dial from 600 to 2600, and Stockfish reviewing the finished game separately once it's over. It starts with a free trial on iOS and Android. It's also a much younger product than either Lichess or Chess.com, with nowhere near their track record or roster size, so if the biggest possible free selection matters more to you than a single dial, Lichess and Chess.com both genuinely beat it on that specific measure.

Whatever bot you land on, the game itself is only half the value: a proper review afterward is what turns a loss into something you actually learn from. We cover what a good post game review looks like, on any platform, in our guide to reviewing your own games for free.

Last verified: 19 September 2026, against the sources cited above. Bot rosters, listed ratings and pricing tiers on products we don't control change without notice; confirm current details on each platform's own page before relying on them.