Fritz 20: A Chess Engine Built to Teach Humans, Not Just Beat Them

2:47 a.m. at a hotel in Da Nang. I reopened the game of a seventeen-year-old player who had just lost in round seven of the national chess championship. His thirty-first move, a pawn push that blocked his own queen and kept the structure intact, I crossed out in three seconds. The analysis engine on my laptop returned an evaluation of minus 1.42. By every objective measure, it was a bad move.

But when I switched to training mode instead of analysis mode, the engine stopped judging. It no longer asked whether the move was good or bad. It asked three different questions: how much time did he have left on the clock, which continuations had he actually calculated, and what happens to this structure if the opponent does not play the best move.

I stayed until nearly four in the morning. The answer was not on move thirty-one. It was on move twenty-two, when he burned forty-one minutes on a decision that should have taken six. The engine told me the move was wrong. It did not tell me why he chose it. The gap between those two questions is the entire story of Fritz 20.

That is why I read ChessBase's announcement of Fritz 20 with more curiosity than excitement. Fritz is not an unfamiliar name. The first version appeared in 2026, developed by Frans Morsch and Mathias Feist and published by ChessBase. In 2026, Fritz 3 beat Garry Kasparov in a blitz game in Munich. In 2026, Fritz won the World Computer Chess Championship. In 2026, Deep Fritz drew Vladimir Kramnik 4-4 in Bahrain. In 2026, in Bonn, Deep Fritz beat Kramnik 4-2, the first time a reigning world champion fell to a machine in an official match.

Thirty years after that game in Munich, Fritz 20 is presented not as a challenger but as a personal trainer. The publisher's message fits into three adjectives: efficient, intelligent, individual. The audience is split into two clear tiers: players taking their first steps into serious training, and players already competing at tournament level.

The Vietnamese context makes this more than a routine software upgrade. We have a generation of players raised alongside software: Le Quang Liem, Nguyen Ngoc Truong Son, Tran Tuan Minh, Vo Thi Kim Phung. None of them learned chess without an engine in the machine. The question was never whether to use one. The question is how.

According to the performance rankings published by the international chess community, today's top engines perform at a level equivalent to roughly 3,600 Elo. The human historical peak is 2,882, set by Magnus Carlsen in May 2026. A gap of nearly seven hundred and twenty points.

That number is meaningless unless you have a way to use it. A machine seven hundred points above you does not teach you chess. It only proves you wrong, and it proves it very quickly. In eighteen years of watching competitive chess, I have never seen a player improve simply because a stronger engine sat on his hard drive.

What Fritz 20 promises to do differently lies in separating two functions that have been merged for twenty years: judging and coaching. Judging says this move is wrong and that one is better. Coaching answers why a player at a specific level chose the wrong move, and how to fix it. Most software on the market only does the first job well.

The Fritz line has kept two features across versions that I consider more important than any search parameter. The first is assisted analysis, where the engine walks through the game move by move and warns you before you blunder, instead of running the whole game and delivering a verdict. The second is the ability to lower the strength of a training opponent. An engine that knows how to play weakly in the right way is more useful than one that plays perfectly, at least for someone learning how to win.

The real value of a training engine is not search depth, but whether it can answer the question: if I am at this level, which move is reasonable? That is a completely different question from which move is best. The machine's best move is often unplayable for a human, because it demands a chain of twenty perfectly accurate moves that nobody sustains for thirty minutes against an opponent applying pressure.

I have written before that no tactic is ever old, only the way we read a game expires. In chess that is true in a stricter sense: same position, same tool, but if the reading has expired, the tool is useless no matter how strong it is. That summer in the V-League I learned one thing: a formation is only beautiful when the opponent agrees to stand still. Engine lines are the same, and in human chess the opponent rarely stands still for more than ten moves.

At Vietnamese youth events I keep observing a paradox that has become a pattern. Players of fifteen and sixteen memorize the first fifteen moves of the Najdorf, yet cannot win a rook-and-pawn endgame against an opponent rated two hundred points below them. They are trained by engines, but only in the opening, the phase where the machine gives the clearest answers and creates the least transferable value. The endgame, where human technique is actually built, is left to instinct.

Data I recorded at a recent national youth event, across forty-two games in the under-fourteen category: thirty-one games passed move twenty without either side deviating more than half a point from the engine evaluation. But only nine of the twenty-eight games that reached a balanced endgame were finished with correct technique. In other words, their openings sit at 2,200 Elo and their endgames at 1,800.

That is precisely the gap a product like Fritz 20 targets. A useful personal trainer has to recognize that the problem for young players is not a lack of opening knowledge, but a lack of endgame technique and time management.

The biggest blind spot of any training engine is not in the algorithm. It is in the user. At the same moment Fritz 20 is presented as a teacher, an engine in every home is also an engine in the hands of every cheat. In 2026, a cheating case at a major tournament in the United States pushed online chess into months of argument over who had used an engine and how. Platforms tightened their screening, but the root cause did not change: a powerful tool released to everyone is released to everyone.

The transfer market is chess, not a card game, yet plenty of sporting directors prefer to flip cards. The chess software market runs the other way: people buy on faith in a brand and on the numbers printed on the box, not on measured outcomes. An engine that is twenty Elo stronger is an easy specification to sell. A feature that cuts thirty percent of wasted thinking time is not.

The second risk is pedagogical. When every game is graded instantly, players slowly lose the ability to judge for themselves. I meet young players who will not play a move without switching on the machine to check it, even in training. They become translators for software instead of chess players. In 2026 I discarded half of my old data set, because post-lockdown football was a different sport. Chess after the engine reached every home is a different sport too, and not an easier one.

Data limits: the information about Fritz 20 in this piece comes from the publisher's announcement and has not been independently verified on the full software build. The 3,600 Elo figure is a reference performance level on community engine rankings, not an official competitive rating, and varies with hardware configuration. The forty-two-game sample from the youth event was recorded directly on site and may be skewed if rapid games not entered in the official record are included.

The question I carry into the next round is not how strong Fritz 20 is. The question is whether a tool designed to teach humans can change how a seventeen-year-old Vietnamese player prepares for move twenty-two. Computers beat humans long ago. What remains is harder: teaching humans to play better than themselves.