FRITZ 20: A Revolution Without a Single Number

There is a professional habit I have kept for nearly twenty years as a data journalist: whenever a product announcement lands on my desk, the first thing I do is count numbers. Not count adjectives. Count numbers. How many verifiable figures? How many metrics with traceable sources? How many quantities that carry a unit of measurement?

This week I opened the promotional material for FRITZ 20, ChessBase's latest chess engine. I counted. Elo ratings listed: none. Benchmarks: none. Release date: not stated. Price: not stated. Technical feature list: absent. System requirements: absent.

In exchange, the document repeats four phrases: "personal chess trainer", "training revolution", "efficient, intelligent, individual", and "for both players taking their first steps and players competing at tournament and professional level".

The ratio of verifiable data to promotional adjectives in this document is roughly zero to one. For a technical product sold to chess players — an audience that reads tournament tables, Elo coefficients and move-analysis charts for pleasure — that is an uncomfortable opening.

Thirty-five years of a single name

Fritz is not a new product. It first appeared in 2026, developed by Frans Morsch and Mathias Feist, and quickly became one of the best-known commercial chess engines of the 1990s and 2000s. ChessBase — the German chess software company founded in Hamburg in 2026 — placed Fritz at the centre of its product portfolio and attached the name to almost its entire engine marketing strategy.

People remember Fritz through three matches. In 2026, Deep Fritz drew 4-4 with Vladimir Kramnik in Bahrain. In 2026, X3D Fritz drew 2-2 with Garry Kasparov in New York. In 2026, in Bonn, Deep Fritz beat Kramnik 4-2 — in the second game Kramnik allowed a mate in one, a moment that entered chess history as proof that a human can collapse under calculation pressure.

I grew up inside the chess world from 2026, when I was both playing and organising tournaments. For my generation, Fritz defined the very idea of a home engine. Not Stockfish. Not Leela. Fritz, with its red and blue interface and the sound of pieces hitting the board.

That is precisely why a promotional document with no numbers feels more irritating than usual. A thirty-five-year-old brand has no need to speak in vague terms.

When engine strength became a free commodity

To understand why the FRITZ 20 material is technically empty, one has to look at how the market structure changed over the past decade.

Stockfish is open source, free, and has long topped most computer rating lists. In 2026, Stockfish 12 integrated NNUE — an efficiently updatable neural network — a technical turning point that let a classical engine learn from a neural network while still running on CPU at high speed. From that milestone, the gap between free and commercial engines almost vanished, and in some formats reversed.

Leela Chess Zero, also free and open source, runs a neural-network architecture on GPU and reaches parity with the leading group across many testing formats. Houdini was the strongest commercial engine from roughly 2026 to 2026 before being overtaken. Komodo followed a similar trajectory.

At the application layer, Lichess offers free engine analysis to every user. Chess.com integrates engine evaluation, lessons, puzzles and automatic game review, and in 2026 acquired Chessable — a structured course platform — then absorbed Play Magnus Group, which included training tools associated with Magnus Carlsen.

Those three currents — the world's strongest engine being free, analysis being free at platform level, and structured training content sitting inside one subscription ecosystem — produce a simple fact: engine strength, treated purely as a commercial attribute, has been priced down to zero.

People called that a shock. I call it unread data.

Against that backdrop, a promotional document saying the product is "more than just a chess engine" sounds like a confession. It is an admission that engine strength is no longer a selling point. The problem is that after the confession, the FRITZ 20 document offers nothing in its place.

What a decent product announcement looks like

I hold one professional rule: before criticising someone for presenting too little data, I must state clearly what data should have been there. If I were ChessBase's chief engineer and I genuinely had a breakthrough in hand, I would publish eight numbers.

One: FRITZ 20's Elo on independent engine rating lists such as CCRL or under TCEC testing conditions. Two: the match rate against the leading engine at equal search depth. Three: ACPL — average centipawn loss — when analysing a standard test suite. Four: computation time on a fixed test position. Five: the number of tablebase positions the engine can access. Six: minimum and recommended hardware. Seven: retail price. Eight: release date.

Not one of those eight numbers appears in the FRITZ 20 material I read.

Let me explain the two most important metrics, because I have never been shy about over-explaining foundational concepts. ACPL, Average Centipawn Loss, measures move quality by taking the engine's evaluation before and after each move, converting the difference into centipawns, and averaging. A player with ACPL 25 plays more accurately than a player with ACPL 60. Engine match rate is the percentage of moves identical to the engine's top recommendation — the number used to compare how closely two engines, or a human and an engine, align. Both are universal standards in any technical report on computer chess.

xG does not replace emotion; it explains why our hearts race. ACPL does the same for the chessboard: it does not remove the beauty of a game, it simply points to where that beauty actually lives. A company selling a training tool without publishing ACPL is asking the buyer to trust the seller's emotions.

Two metrics left blank, and what the blank means

What stands out is not that a promotional document lacks data — that happens daily. What stands out is which data was left out.

Car advertising shows horsepower and torque because those numbers impress easily. Phone advertising shows benchmark scores because buyers have learned to read them. Chess engine advertising shows Elo because Elo has been the currency of this industry for thirty years.

When an engine maker removes Elo from its promotional material, there are two explanations. The first: the number is not impressive. The second: the number did not exist at the time the document was written.

Both explanations lead the buyer to the same place: no basis for comparison, and therefore no basis for valuation.

I wonder whether this document is evergreen content, written to rank for searches containing the product name, rather than a dated launch announcement. The absence of any date means the reader cannot place the document in the right market context. An announcement without a date is an announcement whose timeliness cannot be verified.

The audience problem: first steps to professionals

This is where I paused longest while reading.

A product claiming to serve both players taking their first steps and players competing at professional tournament level is describing two user groups with almost opposite needs. Beginners need concept explanations, graded exercises, simple feedback and tolerance for error. Professionals need search depth, tablebases, multivariate analysis, large opening database management and raw data export.

Chess software market history shows successful products pick one pole. Chess.com owns the mass-market pole with an integrated experience. Chessable owns the structured-learning pole. Stockfish owns the raw-strength pole. Lichess owns the free-and-open pole.

A product standing between all four poles without stating a single technical specification is asking buyers to infer its value. In chess, where users at every level are used to reading metrics, that ambiguity is a competitive disadvantage rather than a positioning strategy.

The Fat Fritz 2 precedent

There is a specific reason I am especially cautious about ChessBase engine marketing documents, and that reason has a date.

In February 2026, ChessBase released Fat Fritz 2, promoted as a neural-network engine developed by the company. The open-source chess community quickly analysed it and argued the engine was a derivative of Stockfish. A public dispute over the GPL open-source licence followed, with Stockfish maintainers raising questions about code use and attribution.

I do not have enough data to judge who was right in that affair, and I will not do so. What I take from that precedent is methodological: when a commercial vendor makes technical claims about an engine without supporting data, history shows the independent community becomes the auditor, and the audit usually finishes faster than the marketing department expects.

The transfer market is the only place where people pay for unverified numbers. The chess software market is the same, with one difference: here, buyers can run the test themselves within thirty minutes of installation. A product that will not show its numbers before purchase is betting that buyers will not check after purchase.

The contrarian angle: missing data is not proof of a bad product

Here I have to correct myself, because this is the rule I set after years of erring in the opposite direction.

No data in a promotional document is evidence about the quality of the promotional document. It is not evidence about the quality of the product. The two quantities correlate in many cases, but correlation is not causation. That is the mistake I have watched colleagues make hundreds of times, and I do not intend to repeat it.

There is a perfectly reasonable scenario: FRITZ 20 is genuinely a good product, and ChessBase's marketing department chose a presentation aimed at casual buyers, on the assumption that casual users do not read Elo. If so, this document is a communications failure, not a product failure.

I also have to concede another possibility: the direction ChessBase appears to be taking — positioning Fritz as an integrated training platform rather than a pure engine — may be the right one. Engine strength has been commoditised, but human learning has not. A tool that helps a player understand why they were wrong, rather than merely stating that they were, still fills a real market gap.

Numbers are asceticism: you must give up what is easy before you can see what is true. But asceticism is not self-flagellation — I am under no obligation to reach a negative conclusion simply because I lack data. I am under an obligation to say plainly that I lack data.

So I set a falsification condition for myself, publicly, so I cannot later revise my words. If within six months an independent test appears — from the CCRL community, from TCEC, or from an independent testing group that publishes its method — showing FRITZ 20 reaching Elo parity with the leading free engines on identical hardware, I will rewrite this entire assessment. If a training feature list is published and demonstrably provides individualised feedback superior to existing platforms, I will rewrite it too. Third condition: if the retail price sits below a standard online course and a trial allows thirty days of verification, I will read that as a signal the vendor believes in its own product.

Three conditions. None currently has grounds for confirmation.

What is actually being commoditised

Age is the one variable that never lies. For a thirty-five-year-old software brand, age says two contradictory things.

The first: the brand has accumulated value. Loyal users, a content library, an enormous opening database, relationships with federations and tournaments.

The second: the revenue structure of the old business model is eroding. The boxed-software model — pay once, use for years — faces direct pressure from subscriptions. Chess.com charges monthly. Lichess is free. Chessable charges per course and, since 2026, sits inside a larger subscription ecosystem.

In that structure, every new version of a boxed product has to justify an upgrade more forcefully than the last. The vaguer the promotional document, the greater the burden of proof.

I have followed this industry since 2026 as a player and tournament organiser, and since 2026 as one of the sources the chess community regularly cross-checks. That experience taught me one thing about software market cycles: products that survive three generations of users share one trait — they publish numbers before asking for money.

Signals for the next cycle

I am not concluding that FRITZ 20 is a bad product. I am concluding that, as of the release of this document, no public data exists on which to evaluate it, and that absence itself is valuable information.

Four signals I will track over the next six months.

First, the official release date and retail price. A product with a date and a price is a product willing to be compared.

Second, independent verification data from community engine rating lists. That is where every marketing claim about technical strength gets tested under public conditions.

Third, a specific training feature list with reproducible examples. If the personalisation mechanism genuinely exists, it must be describable — not with adjectives, but as a process.

Fourth, data on ChessBase's revenue structure over the next two years. If subscription revenue share rises, the shift this promotional document hints at is real — merely presented poorly.

Fans see the goal; I see the pass before it. Here, the pass before it is a promotional document with no numbers. It does not tell me whether the engine is weak or strong. It tells me the writer believed the reader would not check.

For a chess community long accustomed to reading metrics, that is a bet worth watching.