Fritz 20: When the Toughest Opponent Sits Down to Become Your Coach

In Bonn, in November 2026, inside the VTC commentary booth, I sat in front of a screen with a notebook already full of handwriting. The second game of the Deep Fritz versus Vladimir Kramnik match had just passed move thirty-four. On the board, a mate in one appeared like a crack in glass, so clear it was hard to believe. Kramnik, world champion since 2026, missed it. The machine did not. I went silent for a few seconds on air, and inside that silence another thought lit up: a human had just been beaten by the very thing he would later need in order to learn how to stand back up.

Nineteen years later, ChessBase introduced Fritz 20 with three short lines. Your personal chess trainer. Your toughest opponent. Your strongest ally. Read quickly, they sound ordinary. Read slowly, they describe the long road of an industry that finally learned to stop showing off its strength and started teaching people how to be weak at the right moment.

FROM A MACHINE THAT PLAYS TO A MACHINE THAT TEACHES

Fritz first appeared in 2026, developed by Frans Morsch and Mathias Feist and published by ChessBase in Hamburg. Back then the chess world had only one question: can a computer beat a human? In 2026, Fritz 7 drew 3-3 with Garry Kasparov in Bahrain. In 2026, Deep Fritz drew 2-2 with Kasparov in New York. In 2026, in Bonn, Deep Fritz beat Kramnik 4-2. Those three matches closed one chapter and opened another, and I was present in all three through my work as a commentator.

The interesting part is this. Once the machine had proven it was stronger, the question should have stopped. It did not. It changed direction. From “can the machine beat the human” the chess world moved to “what can the human learn from the machine.” That is when chess software changed its nature. It stopped being a sword and started being a forge.

I remember the early 1990s, when I still had to copy games by hand from old magazines to prepare a broadcast. To look up an opening line, you flipped hundreds of pages, cross-referenced from memory, and sometimes waited for a library to open. Today all of that lives inside one program. But speed does not create understanding by itself. That is the beautiful tragedy of our era, and it sits at the centre of the Fritz 20 story.

A necessary technical note: the new generation of Fritz runs on neural-network engines trained on millions of games and on the analysis of the strongest machines. Top engines now sit above 3,500 Elo on independent engine rating lists, a gap no human on Earth can close. But every strong engine shares one weakness: it answers very fast and explains very badly. Fritz 20 is selling a different promise, a promise about the explanation.

THREE ROLES, ONE ARCHITECTURE

The three roles in the Fritz 20 slogan are not three fireworks placed side by side. They form a closed loop, and that loop is the real subject.

Role one, the trainer: the software stores your game history, classifies your mistakes, and builds repetition drills across time intervals. In pedagogy this is called spaced repetition. When I coached junior students in Hanoi in the 2000s, I did this by hand. Each student had a notebook, each week I recorded that student's two worst error types, and ten days later I tested them again. Recall after ten days was far higher with the retest than without it. Software simply does what a good teacher already does, but with a patience no teacher has.

Role two, the opponent: adjustable strength, handicap mode, the ability to play in the style of a specific famous player. This is the most undervalued part of the whole idea. A machine of infinite strength that beats you a hundred games in a row teaches you nothing but despair. Value lives at the border, where you win roughly forty percent of your games, where every mistake has just enough consequence for you to remember it.

Role three, the ally: analysis, blunder checking, cloud analysis, comparing candidate moves. This is the most popular role, and it is also the role that leaves many players standing still for years without understanding why they are standing still.

A chess engine does not teach you how to play chess. It only teaches you how to detect your own mistakes, and most players skip the second half.

Based on my experience covering matches across three decades, I believe a break point appears somewhere around 1,800 Elo. Below that point, better tools produce faster progress. Above it, better tools produce deeper dependence while progress slows along a curve nobody wants to draw.

THE STUDY BEFORE THE BOARD

A friend asked me what there is to write about a piece of software. I told him I am not writing about software. I am writing about ritual.

After my failure at the 2026 World Cup, where I mispronounced an Iranian striker's name three times in one half and then went silent for twelve seconds on air, I set a rule for myself: two hours of research for every ninety minutes on air. I rewatch footage, I cross-check, I look for details the audience never sees but that make the true story. When I look at Fritz 20, I see that same ritual packaged into software.

A serious chess session with software does not begin with the first move. It begins by rereading your own old games. A player must pass through four layers: the result, the bad moves, the reasons behind them, and the mental state at the moment of error. Software handles the first two layers very well. The third it handles only moderately, because a move can be wrong for tactical, strategic, or psychological reasons, and all three look identical on a screen. The fourth layer no software touches, even at version twenty.

This is why I like placing two numbers in the same paragraph. One is the number of games the software can analyse in an hour. The other is the number of minutes you actually spend thinking about a move before opening the software. My three-layer rule is simple: every layer of data must be accompanied by a layer of story, otherwise the number becomes a brick in a wall you will later have to tear down.

THE GAP BETWEEN KNOWING AND UNDERSTANDING

There is a paradox I have observed for more than thirty years: young players today know far more than my generation did, yet they understand less in a few specific regions of the board.

I remember one of my students, fourteen years old, very strong thanks to software. He memorised opening lines twenty-five moves deep, knew exactly which squares were weak, spotted tactical motifs faster than I did. But when the game left the book, when an opponent played an odd move on move eleven, he panicked. Not because he lacked knowledge. Because he had never produced that knowledge himself. Everything he knew had been downloaded.

How does downloaded knowledge differ from knowledge you dug up yourself? It has no roots. In a chess game, at some point the opponent will play a move no software ever taught you, and in that moment all you have left are your roots.

Fritz 20, as a product, is trying to sell the roots instead of the branches. That is why I find the three roles in its slogan genuinely meaningful. The trainer role forces players back to their own old games. The opponent role forces decisions under pressure. The ally role appears only after the other two have done their work. Reverse the order and you get a very strong tool and a player standing still.

The board never lies; we only lie to ourselves with applause.

In chess, applause takes another shape. It is not loud. It is the line reading ninety-two percent accuracy after you analyse a game you lost. It is a beautiful graph of a game you destroyed yourself from move thirty. Those numbers comfort you, and comfort is the first enemy of the serious player.

THE BLIND SPOT OF COLLECTIVE MEMORY

Chess collective memory keeps a lovely image of the pre-engine era: players facing each other, calculating in their heads, decisions made by intuition and experience. That image is true, but it has been romanticised until it hides another truth: most games back then were decided by elementary errors that a twelve-year-old today would not make.

I do not want to return to that era. What I want is to keep some of its slowness. And here is the biggest blind spot of the whole industry: they sell you speed, when what you lack is time.

A new version number every year. A longer feature list every year. A promise that this version will be the last one you need. No version is ever the last one, and that is not the manufacturer's fault. It is the nature of a market. The question is not whether you should buy Fritz 20. The question is whether you have one hour a day to sit still with it.

From keyboard to board, the speed of words is never as fast as the speed of pieces.

If you play below 1,500 Elo, what you need is a teacher who watches you, or a group of friends to play with. A strong program will not help you see your own mistakes, because you do not yet have the standard to judge them. If you play above 1,800 Elo with concrete tournament goals, a good tool starts to matter. The space between those two zones is the land where every advertisement lies politely.

WHEN THE STANDS ARE EMPTY, I HEAR THE POSITION WHISPER IN ANOTHER LANGUAGE

In 2026, when tournaments had to be played in silence, I commentated on several online chess events with no spectators in the room. No applause, no camera shutters, no murmur from the stands. Only the click of a mouse, breathing, and something else it took me months to name.

It was the sound of thought. When the outside goes quiet, you begin to hear what happens inside. This connects directly to how we use software. If you switch an engine on while analysing, you will never hear the sound of your own thinking. You will only hear the answer.

A game with no spectators is still a piece of music, played only for yourself.

This is why I advise serious students to follow one simple law: analyse your own games with your own head first, then open the software. Write down the three moves you doubt most. Form your own hypothesis. Then switch the machine on and compare. At that point the machine is no longer teaching you moves. It is answering your questions. And between those two things lies a gap wider than the gap between two playing levels.

WHAT REMAINS AFTER YOU CLOSE THE LAPTOP

After his defeat in Bonn, Kramnik said the machine understood nothing about chess, it only calculated. That remark was technically correct, and correct in another sense he probably did not intend. The machine does not understand, but it points precisely at where the human understands wrongly. That is a modest function and also the most precious one.

I have lived long enough to see three generations of tools pass through my hands: handwritten notebooks, floppy-disk software, and now neural-network engines on personal computers. Each generation promised more than the last. Each kept one condition for success: the user must be willing to sit still.

Fritz 20 will not make you stronger in the first week. Used correctly, in six months you will notice you calculate more slowly but more accurately, and that sounds like a step backwards to anyone who measures progress in speed. But deliberate slowness is the first sign of real understanding.

If you are considering a new training tool this season, ask yourself one question before you open the checkout page: after you close the laptop, what remains in your head? If the answer is a few memorised moves, you just bought a dictionary. If the answer is a new way of thinking, you just bought a teacher.