Grandmaster Vs. Game Review: Gusev's Legendary Queen Sacrifice

When was the last time you read human-only annotations to a chess game, with zero input from artificial intelligence? Have you ever wondered how a human grandmaster's understanding stacks up next to Chess.com's Game Review, and vice versa?

For this article, I asked our Resident GM Rafael Leitao to annotate a chess game without an engine or even an opening book. The game, Gusev-Auerbach 1946, features my favorite positional queen sacrifice in chess history, and Leitao confirmed that he'd never seen the game before. This way, his analysis was fresh and completely human—and done by a modern grandmaster.

I would recommend reading through Leitao's game analysis first, so that you know what happens in this remarkable game and see the human perspective of a strong player. In fact, we also have Gusev's own notes from the 10/1951 edition of Chess in the USSR. After, you can read the comparison with Game Review's assessment.

GM Rafael Leitao's Non-Engine Analysis:

I've also translated Gusev's own thoughts on the game from Russian into English below. Even though almost a century separates Leitao and the Russian master, their assessments were, interestingly, more or less aligned.

Gusev's Annotations:

In the rest of the article, I unpack the perspectives of human and machine on a deeper level, and I go through the key phases of the game one at a time with diagrams. I share snippets of Leitao's analysis and compare them with Game Review's. 

You can also check out a video NM Sam Copeland published in 2021 analyzing the game, with the help of Game Review.

In general, Game Review was much better at spotting opportunities for counterplay than the human grandmaster. There were several cases where Leitao approved natural-looking, attacking moves, while Game Review found underlying tactical flaws. White's attack looked crushing, but Black had several chances to even hold an advantage.

We will see that even after White's immortalized queen sacrifice, Black had one miraculous chance to save the game—Leitao actually came close to finding it but missed one key move. With the help of engines, we can see that there are often many more opportunities to fight back, or to survive, than we expect. That is, there is great reason to stay optimistic, even if it looks hopeless at times.

One more point. While Game Review might cackle at our inability to calculate millions of positions in a second, I think we can pride ourselves on the ability to explain the critical moments accurately.

That free piece won White a hard-fought game.

—Game Review

The game started off as a Sicilian Dragon, which in itself sets the stage for tactical creativity. With 8...Be6, Black delays castling, and Leitao points out that Black never actually manages to castle in the entire game.

In order to punish this, White goes for a direct assault, with a pawn storm, and this is our first point of disagreement between man and machine. Leitao seems to approve of 11.g4, while Game Review even claims Black is better at this point—something Gusev also underestimated.

To the human eye, White's plan seems logical. Leitao writes:

Brutal chess. White goes for an immediate attack with g4-g5. In the game, Black accepts the knight's return to g8, which seems like a big concession. I'd think of some way to prevent this, taking advantage of the fact that the rook is still on h8.

The heartless oracle finds that it is the white king who is much weaker after this natural move. After 11...Qb6+ 12.Kg2 gxf5 White would face direct problems. 13.gxf5 opens the g-file, while Game Review gives 13.exf5 Ne5 with a large advantage for Black.

The difference in king safety is indisputable at the end of this line (which didn't happen in the game):

White's attack looks overwhelming on the surface, and indeed he wins one of the most spectacular games in chess history. But Game Review points out another moment when Black was better after the what-could-be-more-natural 13.Nd5.

Leitao didn't comment on the move, but once again the fact that Black hasn't castled yet allows him to use the h8-rook in a counterattack. Leitao actually pointed this out conceptually on move 11, but Game Review shows the way with more concrete tactics and variations. 

After 13...gxf5 exf5 14.h6 h4, Game Review says that Black would have been better. However, I needed to see a few more moves before inevitably nodding my head. After something like 15...hxg5 16.hxg5 e6, we can see that the h-file is open for the black rook and that he's challenging the centralized knight on d5.

For even more detail on the position, you can read the full game annotations at the bottom of this article, where I go a little further than that. Black has a significant advantage here.

Black had multiple chances to justify his plan of not castling, but they didn't transpire. As it happened, White went on to win an unforgettable game, and it seems that Leitao could explain the complicated variations better than the computer. Of course, he had the leisure of taking his time, while Game Review returned the analysis in about a minute.

Computers can be annoyingly perfectionistic, and Game Review's comment on 23.Qxd6 isn't that helpful. Leitao doesn't even comment on the move, and Game Review calls it a mistake despite the fact that White's still winning with best play. Better was 23.Qd5 Qe7 24.Rd1 Rd8 25.h3, but if Gusev had played this way, I wouldn't be writing an article about the game 80 years later.

When playing 23.Qxd6 (and possibly earlier than that), White had already seen the combination that would etch this game into chess literature. Leitao writes:

To judge the aesthetic value of a sacrifice, it's important to know whether there was a simple way to win, but I didn't find anything very convincing.

There's no forced checkmate, no direct kingside attack, and yet White wins slowly. Leitao describes this rare situation in his note a move later:

White has almost no material for the queen, but Black's pieces are paralyzed due to the mating theme on f8.

While Gusev himself didn't comment at all on 23.Qxd6 a move earlier, he wrote about his stunning queen sacrifice:

A very interesting situation has arisen. White has a rook and two bishops against his opponent's queen and two rooks. However, Black's rook on h8 isn't in the game, the king doesn't have a single move, and the unprotected white pawn on e6 is equal in strength to a queen.

White will ultimately win the game by pushing his passed c-pawn all the way up the board. Black will be powerless to stop it, with all of his pieces stuck defending checkmate threats on the kingside.

Between strong human chess players, this is probably going to end in a white win most of the time. But Game Review found one miracle save for Black. After 28.Bxc4 bxc4 29.b3, 29...cxb3 works in Black's favor.

After 30.axb3 a5, we would get this position (not in the game):

While Game Review analyzed the non-critical 31.Kg2 Qd8 32.Rf7 and ended its line there, Leitao actually went down the most important line—that is, the line that's most likely to elicit a mistake from a human. Leitao concluded that Black couldn't save it because he missed a single move deep in the line, writing:

To be honest, I don't see much that Black can do other than play Qe7-g7, but the rook ending doesn't seem to give much hope.

After 31.c4 (White threatens to push the c-pawn as in the game) 31...Qe7 32.Kg2, Black has the only move 32...Qa3!! and there's no way for White to improve. The black queen covers c5, the mate threat on f8, and also threatens perpetual checks if the white rook leaves the second rank. 

While Leitao analyzed the line deeply, Gusev didn't consider 29...cxb3 in his notes.

As the game went, White won a unique game, dominating the board with two pieces and a passed pawn. Black's extra queen could do nothing more than spectate the victory march of the c-pawn.

I am extremely grateful to GM Leitao for offering his human analysis to this game. Although his analysis isn't error-free, and nor is Gusev's, it's the mistakes that give us the greatest insight.

Before closing, I've included the full game again but with Game Review's annotations for the critical moments discussed. I've added my initials, "AL," in all cases where I've added something to Leitao's thoughts.

GM Rafael Leitao's Analysis With Added Game Review Lines

What was it like for you to read human analysis without an engine?

Do you know another wild game between titled players that would work for this article format? Share it in the comments below!