ChessBase's FRITZ 20: The Positioning Problem When a Chess Engine Becomes a Commodity

My Bonn file is only three pages thick. The first page is the transcript of game six, 4 December 2026, when Deep Fritz met Vladimir Kramnik in Bonn. The second is a tidy scoreline: 4-2. The third is a handwritten note about a single move that took me two weeks to understand.

Nineteen years later, another document landed in my inbox in Shanghai. It introduced FRITZ 20, ChessBase's latest product. I read it three times, then a fourth. Only on the fourth pass did I notice the most striking thing: there is not a single measurement anywhere in the text. No Elo figure, no calculation speed, no independent benchmark, no release date.

A chess product sold on prose, in a sport that has entered an era where every argument ends with a spreadsheet.

When a chess engine became a commodity

In 2026, in Hamburg, Frederic Friedel and Matthias Wuellenweber founded ChessBase. They sold game databases at a time when every top-level game had to be copied by hand from a magazine into a notebook. Around the same period, a Dutch programmer named Frans Morsch wrote Fritz, and that name quickly became the face of a generation of commercial chess software.

The 1990s and 2000s were the golden age of the boxed-software model. Players bought discs, installed them, and each new version was an event. Fritz held a special place because it was not merely strong; it was the engine behind two cultural milestones. The X3D Fritz match against Garry Kasparov in New York in 2026 finished level at 2-2, and three years later Deep Fritz beat Kramnik 4-2 in Bonn.

I once trusted emotion, until a number knocked on my door at three in the morning.

That number was the percentage of moves a top player matches with the engine's first choice during the opening phase. When I began measuring it across players, I realised that "style" — the thing I had spent thirty years commentating on — was largely another name for a measurable set of error patterns. An engine does not judge style. It counts.

And because it only counts, the game changed. In 2026, the International Computer Games Association stripped Rybka of its titles after finding the engine had copied code from Fruit. In 2026, DeepMind's AlphaZero published the result of a 100-game match against Stockfish 8: 28 wins, 72 draws, no losses. In January 2026, Leela Chess Zero launched, recreating AlphaZero's method in a fully free, open-source project. Stockfish had long been free, open-source, and top of nearly every engine rating list.

By around 2026, raw engine strength was no longer something anyone could sell. Casual players analysed their games for free on Lichess. Serious players paid Chess.com and received an engine, a database, lessons and a community in one package. ChessBase still owned the largest database on earth, but a database was no longer the only place to find the truth about a game.

In that market structure, a FRITZ 20 announcement should have been a technical document. It was an advertisement.

Four gaps in one product pitch

The first absence is time. No release date, no announcement date, no anchor for the reader. For a technology product, this is the most serious omission, because it removes any ability to assess relevance. An engine positioned in 2026 and one positioned in 2026 live in entirely different competitive worlds. Without a date, the reader cannot tell whether this is today's news or a relic of a cycle that has already closed.

The second absence is every technical metric. No estimated Elo. No positions per second. No match result against Stockfish or Leela Chess Zero. Even ACPL — average centipawn loss, the standard measure for judging move quality and the effectiveness of a training tool — never appears. In chess, an engine that publishes no match results is like a player who publishes no games.

The third absence is definition. The phrase "training revolution" is used as a label with no operational description attached. Which algorithm, which method, which training data, what feedback mechanism — none of it is stated. A word like "revolution" only carries weight when there is a specific feature list behind it that a reader can argue with. When there is nothing to argue with, the word becomes noise.

The fourth absence is consistency of audience. The same document claims the product is for ambitious players and professionals, and simultaneously for those taking their first steps. In software, that is the signature of a product that has not found its identity, not of one with genuinely broad reach.

Products that win in today's chess market anchor themselves to a single point and defend it: Stockfish to computational strength, Lichess to free and open source, Chessable to spaced repetition, Chess.com to the integrated experience. A product that claims to anchor to everything is, in practice, anchored to nothing.

A pressing move is not noise. It is a question that data is whispering.

Legacy is not a permanent selling point

The transfer market does not buy the past. It buys what the data has already forgiven.

I have applied that principle to every transfer I have ever assessed, and it holds for software too. Fritz's victory over Kramnik in 2026 was a historic event. In today's valuation, it is a line in the introduction. The memory of an engine that once felled a world champion does not automatically convert into utility for a fourteen-year-old who needs to fix a weakness in pawn structure.

There are players the world forgets, but data never forgets them. And precisely because data never forgets, it grants no privileges to any brand.

What ChessBase genuinely owns is not the engine. It is the database of more than ten million edited games, the opening classification system that the entire chess world uses as a reference standard, and a generation of players who learned the game on its software. Those are genuine moats, unlike engine strength — which anyone can download for free in three minutes.

The paradox is that the pitch loads all its weight onto the asset with no moat, and says almost nothing about the assets that have one. That is a communications choice, and it is also a signal.

The line between coaching and calculation

There is a widespread misconception I encounter in almost every conversation with parents in Shanghai: they believe buying a stronger engine means their child will improve faster.

The data says otherwise. An engine rated 3600 helps a 1300-rated player very little if nobody translates the analysis into a training programme. The value of training software sits in the layer of interpretation, not the layer of calculation. The calculation layer was commoditised long ago.

If FRITZ 20 really is a turning point, the turn must be in moving from "an engine that tells you that you were wrong" to "a system that tells you what to fix next". That is a problem of user data, of individual error history, of classifying mistakes into groups and designing a pathway. None of it is described in the pitch.

Meanwhile, others have already moved into that layer. Chessable turned opening study into a spaced-repetition process with measurable retention. Chess.com folded analysis, puzzles and automated feedback into one continuous flow, then acquired the Play Magnus Group in 2026 to widen the ecosystem. Lichess offers free analysis with a leading engine, charges nothing, and funds itself through donations.

Those three models attack the same gap from three directions. A traditional boxed product that wants to reclaim the position must prove superior value with numbers. The FRITZ 20 pitch offers none.

When vagueness is a choice, not a mistake

Here I have to argue against myself, because that is the only way an analysis avoids becoming an indictment.

Hypothesis one: the absence of metrics is an editorial error. A marketing team unfamiliar with chess writes a pitch for a technical product and chooses emotional language because that is what they know. If true, the problem is the channel, not the product.

Hypothesis two: the absence is calculated. In a market where the strongest rival is free, publishing a lower Elo figure would be self-destructive. Publishing a higher one would be meaningless, because at the top end, Elo differences between engines no longer correlate with real user benefit. When every comparison is either unfavourable or irrelevant, the optimal strategy is to offer no comparison at all.

I lean slightly towards the second, and that is an unexpected conclusion.

Vagueness, in this case, is a rational defensive wall rather than a weakness. It will not fool experts, but it does not need to. It needs to hold a different customer group: those who buy on brand, on habit, on the memory of a name that has been in their home for twenty years.

Here an analogy from the transfer market is useful. A free agent costs no transfer fee and therefore slips past the financial scrutiny a paid transfer attracts, and his hidden cost — wages, signing fees, integration time — only surfaces later. Free engines operate on a mirror logic: they cost no purchase price, but the price is the time you must spend yourself to turn raw data into understanding. Paid products promise to shorten that time. That promise is only worth something when it is measured.

And this is the crux: if FRITZ 20 genuinely shortens the road from data to understanding, that is precisely the number that should have appeared first in the pitch. Its absence says more than any claim.

Signals to track in the next cycle

I do not judge a product by a leaflet. I judge it by what happens after the leaflet.

The first signal is the arrival of independent testing data — from forums, from users who measure for themselves, from methodical reviews. When the first numbers appear, the question shifts from "is this product strong" to "how is this product different from what I already use for free". That is the right question.

The second signal is pricing structure. If ChessBase moves from boxed software to subscription, it is a sign they have recognised that real value lies in continuously updated data rather than a one-time installation.

The third signal is product focus. If future updates talk about learning pathways, error classification, and each user's mistake history, ChessBase has chosen the right battlefield. If they keep talking about engine strength, they are fighting where there is no longer any spoils.

I light a candle for data. But I always let the flame of emotion illuminate the question.

And my question, after four readings of a pitch with no measurement in it, is simple: if a company that owns the largest chess database on the planet cannot produce a single number to prove its product is better, who exactly is going to produce that number?