Unlike other self-learning Neural Network engines, Maia does not search for the best move. Instead it, like human players, makes mistakes which makes it feel like you're playing a real person. But, that real person does not mean a Grand master. You can select the level at which it plays from 1100 to 1900 Elo.
Supposedly, this also helps when analyzing because it better helps you understand the play of those at that level. Whether that’s a valid claim is, I think, open to question. Who would you most benefit analyzing with, a 1200 or a Master?
I think Maia is a great sparring partner because it does seem to play like a human and it's free.
Visit the site HERE.
Note that this is an online program only and in order to use it you will have to click on “Play” and do a one time download that is 90mb in size. It’s a bit disconcerting that I have been unable to locate where or how this download is stored on my laptop.
Here is a 10-minute game I played against it at the 1500 level. The program played a very solid game until move 26 when it blundered and lost its Queen. How close its play was to the 1500 level I can’t say. What I can say is that for practice at the 1100-1900 level Maia does seem like a pretty interesting sparring partner.
[Event "G/10"]
[Site "?"]
[Date "2025.??.??"]
[Round "?"]
[White "Tartajubow"]
[Black "Maia 1500"]
[Result "1-0"]
[ECO "D00"]
[Annotator "Stockfiah 17.1"]
[PlyCount "57"]
[EventDate "2025.??.??"]
{Blackmar Gambit} 1. e4 d5 {A move variously known as the Scandinavian Defense,
Center Counter Defense or Center Counter Game.moves:} 2. d4 {Transposing into
to the Blackmar Gambit in which white intends to play f3 and get rapid
development and active piece play. It;s aggressive, but its soundness is open
to debate. Most experts consider it unsound and probably rightly so.} {If
black does not want to accept the challenge of proving the gambit unsound he
can play 2...e6, 2...c5 or 2...Nc6} dxe4 {If black does not want to risk
facing the Blackmar and accept the challenge of proving the gambit unsound he
can play 2...e6, 2...c6 or 2...Nc6} 3. f3 {Or, 3.Nc3 first.} exf3 {Black can
play 3...Nf6 if he does not want to accept the challenge of proving the gambit
unsound.} 4. Nxf3 {I have seen 4.Qxf3 played, but that seems unreasonable,} Nc6
{The most frequently seen moves are 4...Bg4 and 4...Nf6} (4... Bg4 5. Bc4 e6 6.
O-O Nf6 7. Bg5 Be7 {Deiller,P (2091)-Jeannerot,L (1950) Val Thorens 2009.
Black is a solid P up.}) (4... Nf6 5. Bc4 e6 6. Bg5 Be7 7. Nc3 c6 {Ivanov,
O-Panamski,S Sofia 2008. Black;s position is a bit passive, so white has
compensation for the P.}) 5. d5 {Gaining space and time.} Nb4 6. c4 {The
threat is Qa4+ winning the N.} e6 {A good move...it attacks the white center
Ps and prevebts Qa4+} 7. a3 {An equally good alternative was 7.Nc6, but I
wanted to force the N to a6 where I felt it would be poorly placed and I hoped
to be able to play Bxa6 which I felt would leave black with a serious handicap
in the form of a wealkened Q-side.} Na6 8. Nc3 (8. dxe6 {is inferior. After}
Qxd1+ 9. Kxd1 Bxe6 10. b4 O-O-O+ 11. Kc2 {Black is better.}) 8... exd5 {
At least as good was 8...Nf6} 9. cxd5 {I thought I had a pretty good
positional advantage here, but Stockfish evaluates it as perfectly equal...
black does have an extra P.} Bg4 {[%mdl 8192] This looks reasonable, but it
gets a question mark and suddenly the evaluation puts white at a winning 3
Pawn plus advantahge! Either 9...Bd6 or 9...Nf6 keeps things equal. Now it's
white to play and win!} 10. Bxa6 {According to plan . I did not even consider
other moves. White's advantage is now only minimal.} (10. Bb5+ {Black has two
defensive tries.} Bd7 11. Bxd7+ Qxd7 12. Ne5 Qe7 13. O-O Nf6 14. Qa4+ Kd8 15.
Bf4 {Black is helpless.}) (10. Bb5+ c6 11. Qe2+ Qe7 (11... Be7 12. dxc6) 12.
dxc6 {is winning for white.}) 10... bxa6 11. h3 Bxf3 12. Qxf3 Nf6 13. Bg5 Be7
14. O-O-O O-O 15. Rhe1 h6 16. Bh4 Re8 {This position is dead equal, but the
engine suiggested a plan of doubling Rs on the f-file. I was still hoping to
take advantage of the weak Q-side Ps.} 17. Qd3 Qd6 {Threatens ...Qf4+.} 18. Rd2
Rab8 19. Rde2 {Overlooking the threat! Correct was 19.Qf3} Qf4+ $17 20. Qd2
Qxh4 21. Rxe7 Rxe7 22. Rxe7 {Fortunately no real damage was done by over;
ooking the threat, but now my d-Pawn is a target. I thought the R on the 7th
gave me an advantage, bit Stockfish evaluates the position as a P in black's
favor.} Qc4 23. Qe2 Qf4+ 24. Kb1 Qf5+ 25. Ka1 Nxd5 {[%mdl 8192] Winning the
d-Pawn is a tactical error. Simply 25...Qf4 keeps his small advantage.} (25...
Qf4 26. Qd3 Qc1+ 27. Qb1 Qd2 (27... Qxb1+ 28. Kxb1 {would draw.}) 28. Re2 Qd4 {
with some pressure.}) 26. g4 {It's important to play this before Re8+} (26.
Re8+ Rxe8 27. Qxe8+ Kh7 28. Nxd5 Qxd5 29. g3 Qd1+ 30. Ka2 {Theoretically at
least black has a won Q+P ending,}) 26... Nxc3 {This saves the N, but loses a
lot more.} (26... Qg5 27. Re8+ Rxe8 28. Qxe8+ Kh7 29. Qe4+ {picjs up the N.})
27. Re8+ Rxe8 {No surprise here, but 27...Kh7 would have been better!} (27... Kh7 28.
gxf5 Nxe2 29. Rxb8 g5 30. fxg6+ fxg6 {Black is still quite lost, but he could
have played on especially since this was a 10 minute game.}) 28. Qxe8+ Kh7 29.
gxf5 {Here I took the liberty of resigning for my opponent. Accuracy: White =
69%, Black = 64%.} 1-0
