Stockfish 19(stockfishchess.org) |
Stockfish 19(stockfishchess.org) |
Let’s take the very famous game Kasparov - Topalov, Wijk aan Zee 1999 and look at the position just after Topalov accepts Kasparov’s Rook Sacrifice:
https://samboy.github.io/blog/Kasparov.html#RookSacAccepted - https://lichess.org/lwiPq9wB#48 - White to move
FEN: b2r3r/k4p1p/p2q1np1/Np1P4/3p1Q2/P4PPB/1PP4P/1K2R3 w - - 0 25
Run the above position in Stockfish 18 with a depth of 25: [1] It shows that White is 3.53 pawns ahead—i.e. White is clearly winning.
Run the above position in Stockfish 19, however, and Stockfish 19 at depth 25 says that, while White is ahead, it’s only a 1.2 pawn edge—Black supposedly still has good drawing chances.
At a depth of 35 ply, Stockfish 19 sees that it’s a clear White victory (6.79 pawns ahead), but it takes far deeper search for Stockfish 19 to see the win compared to Stockfish 18.
[1] To run the position, I type the following commands:
position fen b2r3r/k4p1p/p2q1np1/Np1P4/3p1Q2/P4PPB/1PP4P/1K2R3 w - - 0 25
d
go depth 25
Any usability of stockfish evaluation for human analysis is collateral.
Unfortunately there seem to be no serious fork of stockfish that would care about usability for human analysis.
I was recently doing some larger study and I stumbled upon a bug/feature of stockfish where it accepts small number of invalid cache reads for a lot of speed.
It would result in infrequent blunders which however affected my analysis, so I removed this optimization in my fork.
https://lichess.org/analysis/1b3r2/8/4P3/2pP2k1/3nB2p/3P3P/5... (now finds 1...Rxf2 = immediately, the evaluation is closer to 0.0)
https://lichess.org/analysis/8/kp2q3/1Np5/pPPp1b2/P2Pp1p1/4P... (finds 1.Nc8+ = quicker (let it ponder four lines), the evaluation is closer to 0.0)
https://lichess.org/analysis/rk6/1p1b4/pBp5/PPPp4/3Pp3/4Pp1p... (The evaluation is closer to 0.0, though still far away)
https://lichess.org/analysis/7k/4p3/5p2/1p1p1p2/1PpP1PpB/1pP... (Evaluation is closer to 0.0)
position fen 1b3r2/8/4P3/2pP2k1/3nB2p/3P3P/5PP1/5KRR b - - 0 1
d
go depth 35
In Stockfish 19: Black is 1.2 pawns ahead, and it wants to play Rxf2!
So, yes, here Stockfish 18 can’t find Rxf2!, but Stockfish 19 does starting at depth 13.
> https://lichess.org/lwiPq9wB#48
When I go to that link it shows Black is winning with -1.9 eval at depth 32 with SF19 1MB NNUE! It looks like it got this evaluation from the cloud database.
Showing Black winning in that position is clearly a bug; at 20 ply Stockfish 19 sees a +0.3 edge for White (which is also wrong; it’s about +5 for White, i.e. clearly winning) and only a +0.6 edge at 24 ply.
To be fair, Topalov, right after this game (1999, so before modern computer analysis), analyzed the game for hours with his second and thought Black still had strong drawing chances after accepting the rook sacrifice.
https://nanochess.org/archive/toledo_javascript_chess_3.html
I'd rather spend all that time reading up on game theory and play some fun video games, like Red Dead Redemption.
Nocheto https://nocheto.sallyx.org/ (this my favorite)
ChessMD https://chessmd.org/
PyChess https://pychess.github.io/about/
If you’re a Mac user and want something with native look and feel, there is also The Marvelous Brass Chessplaying Automaton (TestFlight beta):
https://testflight.apple.com/join/mqfhpmHY
There are also iOS and watchOS versions, but those can’t run 3rd party engines like Stockfish.
I don't play Chess much any more but others have taken up the Chess DB mantle (with slightly different criteria).
I do find it ironic that a project I did in a couple of days for myself and chucked online in case it was useful is still kicking around though.
If you like En Croissant there is a nice fork called Pawn Appetit that does all the same things and adds some other QoL bits and bobs.
Anyways. Now I can use an even better Stockfish to benchmark my own engine..
Presumably it's easy to highlight ones from benchmarks where the evaluation of Stockfish 19 is most significantly different from version 18?
There was a succession of interesting engines (I think The King used to win a number of computer tournaments, this is the one Chessmaster used, that wonderful audio-visual program, like Encarta but chess related) and then the big three of Fritz (match against Kramnik), Junior (match against Kasparov) and Shredder. And also HIARCS and later, Rybka (and Crafty, as well, touted as the open source chess engine)
Stockfish has been top dog (fish?) for a while now but there was a notable exception when AlphaZero (which later became Leela? I'm not sure) beat it, but these days of course Stockfish is far and away the best engine (I guess. I don't know how far behind Leela is)
And, yes, human vs computer chess has stopped being interesting for a while. There were some centaur chess (human and computer; humans using computers to analyze) competitions in the 2000s and some interesting things but nowadays of course your phone can handedly beat Magnus Carlsen and others without problems, reliably
Pretty much. DeepMind never released their code or models, just the main ideas of their techniques and training.
Leela, which came out of the go-ai world, implemented the paper(s) and became the de-facto version everyone used for a while until other projects improved over it on both go (mainly katago) and chess (mainly stockfish).
Stockfish uses a combination of classic Alpha-Beta pruning along with a type of AI called NNUE to evaluate positions.
see https://tcec-chess.com/#div=p&game=1&season=29
click on the Standings tab to see the results.
(If you ever end up going for it, don't use AI it takes away a lot of the fun)
The process is very rewarding and there are great resources (chessprogramming wiki). To get an initial bot play random moves is not that much work but it is a good starting point because you can hook it into a gui and watch.
From there you do the move generator and the search and then optimization possibilities are endless :)
Something as simple as encoding the 64-square board in a 64-bit number leads to a rabbit hole that can last years.
After that point, the "AI" part is already solved and you just need to implement it (alpha-beta + all the heuristics).
Most of the recent chess engines simply implement a neural network (by copying Stockfish) and suddenly are at 3000 elo with no effort! Such a waste.
Even if the opponent is the latest Stockfish with 1000x more time per move, it will always be a draw.
I asked about this a few years ago in a Stockfish Discord and the above claim is what someone implied, if I remember correctly.
Stockfish has been dominant for a long time. And a later version of Stockfish would outperform an earlier version given the same time control.
Basically this optimization means that stockfish will sometimes take a bad read from a cache (so it will take an evaluation of the position from cache which was calculated for a different position). It's fairly random when this will trigger a bad play, but if you brute force vs any deterministic setup you will find it eventually.
But even withtout it - you can just implement a kind of a brute-force heap of positions using stockfish's own eval and you can find wins from almost any position even on fairly high playout count.
I got as far as 32M from different even starting positions, but I spent compute to find as many games as possible rather than as deep as possible.
Example game: https://lichess.org/XVUjezMr#134
the collection of the games I found (vibecoded gui) https://kuboble.com/chess_ideas/
> against Stockfish 18, this new release brings an Elo gain of up to 44 points, and wins more than three times as many game pairs as it loses.
https://youtube.com/playlist?list=PLiZ56cRXeIu3-zr-S7_XND3M3...
GM Mathieu Cornette is an excellent chess analyist, might not be the best at banalazing, but he does find some amazing computer games. Engine games are often boring, so it's nice that they pick the good ones.
I wish these engines could explain why it is recommending certain move, or certain sequence of moves, using natural language - it would be very useful for learning.
This is nice to see. One of my few gripes with NNUE engines is their huge size compared with the old “HCE” engines I was accustomed to (which were usually hundreds of KiB or low single digit MiBs in size). I will trade off size for superior playing strength but at a certain point it becomes hard to keep justifying a 2x increase in size for what seems like ever diminishing practical returns. I am still searching for a strong NNUE engine that is ~10MiB or less. I have found Marvin, which is small at 1.8 MiB, but it is nowhere near as strong as Stockfish.
"The ability to play chess is the sign of a gentleman. The ability to play chess well is the sign of a wasted life."
I find this to be very valid.
I wonder if anyone at those companies is trying to tackle this problem.
I'd argue that as we condition ourselves into a state of constant development, we shouldn't even call these things "tools" anymore, but pieces of an organism that spurn us to upgrade and complete it at every step.
Tools working for us only applies to local, simple tools that one person or a small communiity can basically understand and make, like a hammer. That's a tool. I wouldn't call a computer a tool any more just like I wouldn't say that our brain is merely a tool. It's more: it's an extension of us and even an integral part of us that modifies our very instincts and desires itself.
When you're amazing, you understand what you're doing and where to improve. For very complex systems, the improvements will be in confined areas (not rewriting everything but noticing a bug or reading about another project which found that some other hyperparameter works better). Or even if you have no idea, the simplest way would be making random tweaks and seeing which ones are advantageous
This was all I thought of when I read "Stockfish"
The results are absolute garbage (as you would expect) since the LLM's really don't understand chess or how to play it, someone posted a benchmark of the current SOTA ones playing Chess and the best was estimated at 1480.
Other aspects of strategy, like shortening decision loops[1], forming alliances, or the exploration/exploitation tradeoff[3], are simply not represented by chess because it is a turn-based, zero-sum game with perfect information. That makes chess a poor model of real-world strategy.
Still, the things it does teach are real and useful, so as long as you don't think its the end-all, be-all of strategic thinking you can get something out of it.
[1]: https://en.wikipedia.org/wiki/OODA_loop
[2]: https://en.wikipedia.org/wiki/Cooperative_game_theory
[3]: https://en.wikipedia.org/wiki/Exploration%E2%80%93exploitati...
It also teaches you to forget about the past and focus on the present. It teaches you long term plans usually don't work and relying on them is a bad idea and it's better to focus on execution here and now (this is what Soviet school got wrong although probably only officially, the players knew it all along - analogies to 5 years plans were just too tempting for book authors).
At human level it also teaches you to design strategy to exploit your opponents - play openings they hate, simplify vs players who like complicated positions, complicate vs players who like simple ones.
There is a lot of interesting things you learn about thought process and decision making as your level improve. It might not be the best game for that but it's pretty good considered how simple it is.
That is true for an engine.
Meta-games around studying your opponent, specific match preparation and all those game theoretic stuff are so pervasive people claim they are 'killing' chess at top levels. Anyway, for humans, chess at high levels of mastery is a striking display of bootstrapping some insane cognitive machinery around memory, processing speed, 'expanding' working memory via understanding:
[0] Alekhine blindfold simultaneous vs 32 opponents: https://www.blindfoldchess.net/assets/images/alekhine-excerp...
[1] Reingold, Charness, Pomplun & Stampe (2001), "Visual Span in Expert Chess Players": https://pubmed.ncbi.nlm.nih.gov/11294228/
[2] Connors, Burns & Campitelli (2011), "Expertise in Complex Decision Making: The Role of Search in Chess 70 Years After de Groot": https://onlinelibrary.wiley.com/doi/epdf/10.1111/j.1551-6709...
Off the top of my head I can think of a number - the value of diversity and the contextual power of capabilities such as the power of weaker pieces used diligently, and the power of counterintuitive sacrifices.
Chess also makes no accommodation for your fantasies and ego. There is no luck, only skill.
There is also much to be garnered from the layered complexity of strategy, where some strategies will take you to a certain level of play that need to be completely replaced or expanded with nuance to reach a higher level.
There are many lessons to be learned from chess.
That said, I would not recommend anyone become good at chess to get good at life.
There is a very tangible benefit to people playing in chess that has nothing to do with the game itself. One example is for kids - I think kids playing chess (particularly over the board) learn quite a lot from the game. For example, etiquette (being respectful to your opponent), patience, logic, etc… but one of the best things they learn is how to lose. In chess you lose a lot. Asymptotically vast majority of players will lose approximately 50% of their games. Learning how to lose (respectfully), learn from your mistakes, then overcome them is really great process to learn. Doing this in real life rather than over the screen is also completely different.
Combine that with the game being fun and the social aspects (meeting new people, making friends)… chess very much transfers to real life in my view
That's valid for pretty much every kids sports/games though
You're so evil lol.
And that's when I realized nah, it's not worth it.
That's a great quote, thanks for sharing.
It can be a bit fun, sometime, to play someone that has studied an opening at my level, but since they aren't that deep into it, they only know the moves for some variations. So me, not knowing the opening, do something from the left-field, and they don't really know how to punish it anyway. Remember one time someone got a bit grumpy, heh.
That's perhaps true for full information, two player games.
Try Diplomacy for a change. (Even if you play without explicit negotiations and only have the moves speak for themselves.)
Some folks find chess fun and interesting enough to play it and study it all their lives. Other games can also be fun, though it's rare for a game to hold someone's attention throughout their whole life.
So, while I did play chess for fun, after a while it became boring--but I also impute this to my ADHD, perhaps for others it might not stale.
I would not mind a game or two with an interesting opponent, but playing daily is out of the question for me anymore. Time has opportunity cost, right?
- dealing with imperfect information
- bluffing
- thinking ahead with an enormous number of possible moves (especially when you consider deck construction)
- tracking board state
- mentally implementing "triggers"
I hit about 2k Elo on lichess playing the 3+0/5+0 and I have no regrets!
To each their own.
If you like the competitive aspect of chess, you may enjoy physical sports too. I started to find video games pretty boring compared to pushing myself physically.
Nassim Taleb say a lot on this in his Anti-fragile.
It's just a board game
- it teaches resiliency in bad situations.
- it teaches you to not be too harsh on yourself as a person. If you watch GMs playing online, they make bad mistakes all the time too - they're not godlike creatures who make no mistakes, far from it.
The flipside is that if you approach it the wrong way (I know I have at many points), it can have the opposite effects too.
To approach chess in a mentally healthy way was a long process for me. You need to get rid of your ego and be objective as much as possible when playing, or it will just make you sour - this in itself is a valuable lesson.
Yes, I agree.
Not that an engine evaluation will always be helpful (the line in question making a position strong or weak might be effectively unplayable/impossible to find by a human), but it’s useful for studying nevertheless.
Leela Zero tried to open-source that work, but Stockfish incorporated a number of improvements from AlphaZero, including a neural network and a different search method, and consistently beats Leela Zero.
I have a book, "Game Changer", in which a chess expert calls out several instances where AlphaZero made moves surprising at the time; situations where all chess engines rated things one way and AlphaZero rated them differently. When I enter them into Stockfish now, it usually rates things more similarly to the way AlphaZero did, and often chooses the move chosen by AlphaZero.
The only real test of course would be to dig up AlphaZero and run it again; but I think based on the evidence we have, Stockfish of 2026 would probably trounce AlphaZero of 2018 with equivalent compute available.
It still exists, but it's private / internal, and sometimes used for a few different things.
It was used by Kramnik to test the hypothesis whether no-castling chess was viable (basically chess, but disallowing castling). That was a year after DeepMind published the match they ran of AlphaZero versus Stockfish.
It was still in use last year, I remember seeing some Grandmasters with interests in chess studies were invited by DeepMind to judge the beauty of chess problems composed by AlphaZero (or whatever form the thing that used to be AlphaZero is now).
So it's still kicking around in the background.
There are a few yt content creators that cover great engine games, if you're curious.
I ran fishtest on an old cluster for a while. Was nice to watch the stockfish team methodically improving things. Basically, every commit has to show it can beat the main branch before it gets merged.
What's funny is that if computer chess had paid more attention to computer shogi, Stockfish could have already been using NNUE by that time, and not lost to AlphaZero. But perhaps that would have stifled Leela's development, and would have led to worse training data for Stockfish in that parallel universe's present day, making it weaker overall.
Unknown, it was only around for a little while however there is Lila which is based on the same ideas as AlphaZero (and also a fantastic project generally) they generally trade blows with Stockfish coming out very slightly ahead.
That said Stockfish also uses an NN (NNUE) these days so the line is getting more blurry.
It might help young players learning without a coach, but my understanding is that most are just looking at engine evaluation anyway, as tactical speed is a typical advantage of the youth.
Also - it's not only my change is making stockfish not blunder. Any little change in the code or execution will butterly-effect into no blunder.
To reproduce a bunder you have to be very deterministic. You need to specify threads = 1, and fixed cache sizes.
One example move was 48. g4 from this game: https://lichess.org/osSoJGzO#95
Chess is the first board game you can introduce to your kid that has no element of luck whatsoever. It could be brutal. I have seen 7-9 year olds coming out crying after a game in a tournament. But it also may be requires very good parenting to handle all the emotional distress. Chess also is obviously accessible.
It is an emotional rollercoaster to engage kids in Chess tbh. We are ready to sometimes throw the towel every other week. :-)
I figured it would just be a temporary name until I came up with something better.
Then it stuck. I will say it has been easy for my customers to remember the name!
Chess is a great game for removing all those external factors. It's just thinking
Not everything has to be an absolute competition-or-not. For the same reason you can enjoy a casual game of chess with a friend, you can enjoy a casual game of some RTS. Helps that there are a lot of open source RTS communities that each have their own skills-emphasis.
As originally suggested, when running modern Stockfish at 60 seconds per move the games overwhelmingly tend to draw. Even when the older engines make slightly worse moves.
(Eg limit yourself to commands you can send to vim as you write your Lean proof or whatever.)
crazy to see 110mb.com links in 2026 (on the github repo project), it's been a while
Viable in what way? That it's advantageous to never castle if an engine learns to play with that directive? Or that it still makes for a fun game with that new rule?
Go, on the other hand has a nice handicap system (which doesn't change the nature of the game too much). One can have a fighting chance against a much stronger player with enough handicap (Upto 9 stones). In Chess terms, an IM can have a _decent_ game against a GM. Probably lose. But the GM has to work for it.
It's been fun viewing for me and my chess-interested 9-year-old.
So maybe it matters like that? Just knowing that an oracle like stockfish is saying "this is the best move" may trigger pathways in your brain that promote understanding?
Actually, there's this interesting snippet from the release page:
> These techniques have been applied to hundreds of billions of training positions, all of which have been consistently rescored using a strong Leela net.
So Stockfish's neural network evaluator is actually trained using Leela Zero.
In chess, depth usually wins because of how narrow the search tree is compared e.g. to Go.
