FRITZ 20: A Chess Engine Searching for Identity in the Age of Free Analysis

The digital clock flipped from 01:14 to 01:15. On the third floor of a chess training centre tucked into an old lane in Jing'an District, Shanghai, only two people remained: a thirteen-year-old boy and a coach past fifty. Outside the window, the city was still bright. Inside the room, there was only the sound of a mouse.

The boy rewound the game, stopped at move twenty-two, and turned to his teacher with a question I have heard thousands of times in more than thirty years beside a chessboard: "Why is this move bad?"

The screen answered with a dry evaluation. +0.47. Then a green arrow pointing toward the correct move.

The boy nodded, wrote it down, understood. And understood nothing at all.

I sat at the back of the room, in the shadow of the spare chairs, taking notes. I was looking for the answer to a different question. What is left for a chess engine when every chess engine has become free?

A few days earlier, my inbox had received promotional material for FRITZ 20, the chess software package from ChessBase. The document called it a "training revolution" and promised to accompany players from their very first steps to the professional arena. Beautiful words.

And not a single metric.

I read it three times. I looked for engine strength. Nothing. I looked for a comparison table against rivals. Nothing. I looked for a release date. Nothing either.

That is why I sat there at one in the morning, listening to the mouse and asking myself: if one of the oldest names in computer chess no longer wants to talk about its strength, what does it want to talk about?

To answer that, you have to go back to Bonn.

On 25 November 2026, at the Bundeskunsthalle in Bonn, a Russian man in a white shirt sat down opposite a computer. Vladimir Kramnik, then world chess champion, played six games against Deep Fritz. The prize fund was one million US dollars, with Kramnik guaranteed at least five hundred thousand simply for sitting down. It was the most heavily funded Man versus Machine match ever staged, and it took place exactly nine years after Garry Kasparov fell to Deep Blue in New York in May 2026.

In game two, Kramnik had a drawn position. He only needed to hold. Then he played a bad move and allowed a mate in one. The hall went silent. On the board, there was no prodigy, no trembling human hand missing a chance — only a machine that had seen what the world champion had not. Deep Fritz won the match 4-2, with two wins and four draws.

That was the moment Fritz entered collective memory. For many of us, computer chess has a face, and that face is called Fritz.

Fritz was not born in Hamburg. It came from the hands of Frans Morsch, a Dutch engineer, and Mathias Feist, a German programmer. The first version appeared in 2026, precisely when personal computers began to be strong enough to play chess respectably. In 2026, a version of Fritz caused an uproar by beating Kasparov in a rapid game. In 2026, Deep Fritz drew 4-4 with Kramnik in Bahrain. In 2026, a variant called X3D Fritz drew 2-2 with Kasparov in New York.

ChessBase, the company behind all of this, was founded in 2026 in Hamburg by Frederic Friedel and Matthias Wüllenweber. A year later, the ChessBase 1.0 database software appeared and permanently changed how humans store, search and study chess. Before ChessBase, a coach wanting to check a variation had to leaf through hundreds of pages. After ChessBase, he typed a few characters.

For nearly four decades, that ecosystem — the Fritz engine and the ChessBase database — was the standard kit of anyone serious about chess. In Vietnam in the 1990s, when I was still commentating on chess for national television, a ChessBase floppy disk was passed from coach to coach like a relic. I remember a technician in Hanoi spending two days installing the software on an old desktop; when the database screen finally appeared, the whole room stood up and applauded.

That was the old world. The new world is not so generous.

Today, a chess player in any Vietnamese province, with a mid-range phone, can reach Stockfish for free — an open-source engine stronger than any computer that ever beat Kramnik. Stockfish grew out of a small project called Glaurung, started by Tord Romstad, later taken up by Marco Costalba and Joona Kiiski, and eventually turned into a community of hundreds of volunteer programmers. Nobody pays them. Nobody owns it.

Then, in December 2026, DeepMind published its AlphaZero paper. In a recorded experiment, AlphaZero played one hundred games against Stockfish, winning twenty-eight, drawing seventy-two, losing none. More remarkable than the score was the way it played — committing to sacrifices no engine had ever dared consider.

In 2026, Leela Chess Zero arrived, bringing neural networks into open source. In 2026, Chess.com completed its acquisition of the Play Magnus Group, absorbing Chessable, a training platform built on spaced repetition. Lichess remains free, open source, and ad-free.

That picture reveals a harsh truth: analytical strength has become a commodity. Nobody can sell what a rival gives away.

Meanwhile, on the other side of the world, Chinese chess was moving in the opposite direction. On 30 April 2026, in Astana, Kazakhstan, Ding Liren beat Ian Nepomniachtchi in the rapid tiebreak 2.5-1.5 after the two finished level at 7-7 in classical chess. Ding Liren became the first Chinese classical world chess champion. I was in a live viewing room in Shanghai that night, and I will never forget the sound of it: before the tiebreak ended, the room was utterly silent; after the final move, applause erupted as if at a football match.

What I heard in that training room at one in the morning, and what I read in the FRITZ 20 promotional document, are two ends of the same story. At one end, a chess culture hungry for new tools. At the other, a long-established toolmaker struggling to name itself.

The FRITZ 20 document makes three large claims. First, the product is "a chess engine" in the technical sense. Second, it is "more than an engine" and constitutes "a training revolution". Third, it serves two user groups simultaneously: players taking their first steps, and players already competing at tournament level, including ambitious amateurs.

Placed side by side, those three claims do not form an argument. They form three gaps.

The first gap lies in the word "engine". A chess engine is assessed by specific metrics: Elo strength on engine rating lists such as CCRL or TCEC, calculation speed in kilonodes per second, compatibility with common engine communication protocols, and machine requirements. The document names none of these. For a product positioned as an engine, this is a conspicuous silence.

The second gap lies in the word "revolution". Revolution is a verifiable concept: it requires showing what changed, how it changed, and why users should care. A training revolution should come with a description of method — how the curriculum is organised, which algorithm adjusts difficulty, what data generates a personalised path, which mechanism measures learner progress. The document offers no technical description of any kind.

The third gap lies in the word "audience". In commercial software history, winning products are usually those bold enough to say "I do not serve everyone". Photoshop does not teach drawing to toddlers. Stockfish has no beginner mode. Chessable does not try to be an analysis tool. A product claiming to serve everyone from those placing their first pieces to those competing for a national championship place is signalling uncertainty about identity, not breadth of engineering.

The core insight sits here: in a market where the strongest engine is already free, the value of a commercial chess product no longer lies in computational strength but in curriculum — in the ability to turn a soulless evaluation into a meaningful lesson. And the FRITZ 20 promotional document, as it stands, has not demonstrated that the product possesses that value.

To understand why this matters, look at how training metrics actually work.

ACPL — average centipawn loss — is the standard measure of a player's move quality. If a 2026-rated player has an average ACPL of 22 across a game, it means every move, on average, surrenders twenty-two per cent of the advantage the engine considers optimal. That tells you a great deal about the player, but nothing about what the player should practise tomorrow.

Similarly, engine match rate — the percentage of a player's moves matching the engine's top recommendation — is useful for comparing games, but it does not generate a roadmap.

What I would want to see in a document about a "training revolution" is material of that kind: a classification of learner errors by game phase; an algorithm identifying recurring tactical weaknesses; a mechanism building exercises from the user's own games; an interface showing the gap between opening knowledge and middlegame execution. None of it appears.

Of course, I must be clear: absence from a promotional document does not mean absence from the product. FRITZ 20 may well contain those features. But that is precisely the problem — for a journalist, and for a buyer.

Based on my experience tracking chess matches and tournaments for more than thirty years, I have learned a simple rule: when a manufacturer does not talk about performance, it is usually because performance is not their strength.

And here is a detail anyone reading carefully will notice: the source of this document is the manufacturer. ChessBase is promoting FRITZ 20, made by ChessBase, in language written by ChessBase, to sell to ChessBase customers. This is not wrongdoing — every company does it. But it raises a question about information value: readers are receiving a commercial message and should know they are receiving a commercial message.

In the chess world, this has a memorable precedent.

In 2026, when AlphaZero was widely publicised, many media outlets reported that a computer had "taught humans to play chess". Leading players reacted in the opposite way. They recognised that AlphaZero's plans were too difficult for humans to execute under real tournament conditions, because they demand a level of calculation and a tolerance for risk no one can sustain across six hours. The machine showed humans a door, but the door was higher than human reach.

That lesson applies directly to the FRITZ 20 story. A powerful tool does not automatically become a teaching tool. An engine can tell you which move is best, but it cannot tell you why you keep missing that move across ten consecutive games.

The distance between "knowing" and "being able to do" is the distance an entire chess software industry is trying to close. And this is where the story becomes interesting, because the market already has a winner on this side.

Chessable built its entire business model on spaced repetition, borrowed from cognitive science. You do not read a lecture on the Spanish Opening and go to sleep. You meet that variation again on day one, day three, day seven, day fourteen, until it becomes reflex. In a market where every engine is equally strong, the winner is whoever understands how the human brain remembers.

So if ChessBase is moving toward that model — which the phrase "training revolution" suggests — then what is notable is not the product but the shift. A company that once sold the strongest engine in the world must now sell… something else. And the fact that the document cannot yet name that "something else" concretely is the most analysable datum in the whole story.

By information standards, the FRITZ 20 promotional text contains almost no verifiable facts. No price. No launch date. No minimum system requirements. No concrete feature list. No quantitative claim about training effectiveness. All it offers is a string of adjectives outbidding one another: efficient, intelligent, personalised, comprehensive, revolutionary.

For a software product, this is an unusual approach. Software is bought because it does something. If a company cannot say what it does, buyers must guess — and guessing in a market where the free alternative is always within reach is a dangerous game.

But I do not want to stop at criticism. Criticism is the easiest part of this job.

What troubles me more is the reverse question: if the document is so weak, why does it exist? Why would a company with nearly forty years of history choose to speak so vaguely?

There is an answer I believe is correct, and it is far more beautiful than the criticism.

The collective memory of chess has bound Fritz to a single moment: Bonn, 2026, when a machine felled the world champion. We remember Fritz as a conqueror. And because we remember it that way, we still measure Fritz with the yardstick of 2026 — with the question "how strong is it?".

That is the blind spot.

That question has been obsolete for at least a decade. Once Stockfish reached a level of strength far beyond any human champion and was given away free, whether another engine is a few dozen Elo stronger or weaker no longer generates commercial value. The war over Elo is finished. The winner was the open-source community, and the prize was… no prize at all, only the truth.

What remains to compete over is something entirely different: the ability to lead a human being from knowing to doing. That is an educational problem, not a computational one. And in education, the advantage does not belong to whoever calculates fastest.

Reading the FRITZ 20 document through that lens, its vague phrases take on a different meaning. "Personal chess trainer". "From first steps to professional level". "Training revolution". That is not the language of a company hiding weakness. It is the language of a company trying to move between two identities, and not yet finding the words for the new one.

A powerful machine is easy to describe. You simply say: it is stronger than the previous version. A curriculum is far harder to describe, because it depends on pedagogical philosophy, on a view of how humans learn chess, and on choices a designer must make among hundreds of possibilities.

There is one thing I am certain of, after all these years sitting in rooms like tonight's: what will constitute the value of chess software in the coming decade is not engine strength, but the honesty of the curriculum — and honesty cannot be advertised with adjectives.

In the Jing'an training room, the coach past fifty did something I found more instructive than any marketing claim.

He opened two windows on the screen. On one side, the paid engine installed on the centre's computer. On the other, Lichess, running in a browser, free. He showed the boy the same position in both windows, pointed to identical evaluations, and said a sentence in Chinese that I translate as: "This does not teach you. This only speaks. What teaches you is the teacher, the books, and the games you lost and never forgot."

Then he took out a stack of paper. On it were twelve positions printed by hand, each with a question he had written himself: "If you were Black, would you dare sacrifice this knight? Why?" There was no button to check the answer. Only a pencil and hesitation.

The thirteen-year-old sat motionless before the fourth position for a long time. He chewed the pencil. He erased. He wrote again. And in that moment I saw something no engine can produce: a human being slowing down.

That slowing down is the most valuable thing in sport. It is also the hardest thing to sell.

Over thirty-seven years watching this industry, I have seen many revolutions announced. I have seen software promising to change how children learn chess, then vanish within three years. I have seen engines advertised as the strongest on the planet, then overtaken by a free project within a single season. Each time, I wrote a piece. Each time, I reminded myself that my job is not prediction, but recording what is happening before it has had time to become legend.

And this time, what is happening is a silence.

A silence between a long-established company and a market that has changed. A silence between an undated document and readers who can verify anything in three seconds. A silence between the question "how strong is this software?" — which no one should ask any more — and the question "what does this software teach?" — which no one has answered.

An empty stadium still holds the applause of those who have gone. In this case, there is no stadium. Only a training room, a table, two windows on a screen, and a boy trying to decide whether he dares sacrifice the knight.

When I left the centre at nearly three in the morning, Shanghai had grown colder. On the way home, I thought about what a manufacturer could do if it were genuinely confident.

It would publish metrics. It would state clearly who the product is for. It would let people try it and judge for themselves. It would not call anything a "revolution", because real revolutions do not need to name themselves. Real revolutions leave traces in training rooms at one in the morning, when a child suddenly understands something he will never forget.

After three decades holding a pen, I realise I have never written about sport. I have only written about the moments people choose the harder thing.

And the harder thing, right now, for a chess software company with nearly forty years of history, is not producing a stronger engine. The harder thing is daring to say what it is selling, when what it used to sell has become free for everyone.

That night, on the way home, I wrote a single line in my notebook: "Monitor. No conclusion yet."

In my profession, that is usually the most honest line. This time, I think it is also the most predictive. Because over the next twelve months, what is worth watching is not a software version, but whether a long-established manufacturer dares to replace adjectives with data.

If it can, chess will gain another tool. If it cannot, chess will lose nothing — because the children of Jing'an will still be sitting there, with pencils and questions that have no answer-checking button, teaching themselves to slow down.

And perhaps that is the most encouraging thing in this whole story.