FRITZ 20: When Chess Software Is Sold With Adjectives Instead of Specifications

In November 2026, in Bonn, Vladimir Kramnik sat in front of a screen and lost to Deep Fritz by the score of 4-2. It was the first time a reigning world champion fell to commercial software that anyone could buy and install on a desktop computer. Nearly two decades later, I opened the promotional text for FRITZ 20, the direct descendant of that machine. The document is long and smooth and dense with adjectives: training revolution, personal chess coach, efficient, intelligent, individualised training. Counting through the whole text, the number of technical specifications offered is zero. Not one engine-strength figure, not one comparison with a rival, not one price, not one release date, not one system requirement. For someone who has spent most of his career sitting down to read a data table after each game, that silence is more notable than any advertisement.

ChessBase is the German chess software company founded in 2026, and Fritz emerged from the work of Frans Morsch and Mathias Feist. For more than a decade, Fritz was the strongest commercial engine money could buy. In 2026, Deep Fritz drew 4-4 with Kramnik in Bahrain. Four years later, in Bonn, Deep Fritz won 4-2. In that period, raw computational strength was a competitive advantage you could measure: people bought Fritz because Fritz was stronger, and they knew it through publicly organised engine championships, through score tables, through matches with arbiters.

Then the market turned. In 2026, Stockfish appeared as open source, descended from Tord Romstad's Glaurung, and free. In 2026, Leela Chess Zero arrived using AlphaZero-style reinforcement learning, also free. Lichess offered analysis tools at no cost. Chess.com folded analysis directly into the playing and learning experience. Within fifteen years, the thing that had once been the most valuable asset of a chess software company — pure computational strength — became a commodity available at a price of zero. That is the period you must use to read any press release about FRITZ 20: read it as a statement of market strategy, not as a specification sheet.

Read closely, every information point in the document belongs to the same category: unverifiable qualitative assertion. FRITZ 20 is a chess engine. FRITZ 20 is more than a chess engine. FRITZ 20 is a training revolution. It is for ambitious players and professionals competing at tournament level. It is also for players taking their first steps. There is no operational definition for any of those concepts. There is no algorithm, no training methodology, no exercise structure, no feature list.

The most striking element is the pairing of target audiences. On one side, the product aims at tournament-level players and professionals. On the other, it aims at someone learning their first move. As marketing, that is a coverage strategy. As product design, it is the signature of a product that has not decided whom it serves. The chess software market currently runs in the opposite direction: Stockfish dominates pure engine strength, Chessable dominates structured learning, Chess.com dominates the integrated platform experience, Play Magnus once held the guided-training segment. A product that wants to serve both beginners and professionals must prove it through structure, not through slogans.

If FRITZ 20 really is a turning point in training, there are numbers that could be published without revealing any secret. For example, the average reduction in ACPL — average centipawn loss per move — across a user group after one hundred training games. Or the rate at which moves match the engine's top recommendation, broken down by rating band. Or the retention rate for opening theory after thirty days. Or something simpler: minimum system requirements, disk footprint, the number of tactical exercises, the number of sample games bundled in. None of those items appears. When a manufacturer stops publishing specifications, the silence itself becomes a specification.

Based on my experience tracking games, I keep my own spreadsheet for every major game since 2026: the average deviation between the move played and the engine's top move, the thinking time for each move, and the conversion rate of an advantage into points. Everything on the board is data waiting to be read — if you are willing to sit down. In that spreadsheet, what I learned was not which engine is strongest, because at this point they are all strong enough for any competitive purpose. What I learned was at which level a human begins to imitate the machine, and at which level a human can still beat the machine by understanding its own margin of error. A training product has value precisely in the space between those two points.

Training is a category sold on faith, because its results only appear months later, and nobody comes back demanding a refund after six months of practice without improvement. That delay is ideal ground for marketing language. A stronger engine can be verified in an afternoon with a handful of games. A better training method cannot. This is why makers of training software have a clear incentive to talk about transformation rather than mechanism, and it is also why buyers must demand evidence for themselves.

In software economics, when a component becomes free, value migrates up the stack: to workflow, to user data, to daily habit. The engine is raw material. What is expensive is the way it is organised into a feedback loop with a path through it. Magnus Carlsen does not need an engine stronger than Stockfish; he needs a process that filters what is worth learning out of millions of lines of analysis. So does Alireza Firouzja. The value sits in the filter, not in the calculation.

ChessBase holds an advantage its free rivals do not: an enormous game database, a customer base of serious tournament players, and a brand tied to the era when chess software was a professional tool. But distribution advantage does not automatically convert into product advantage. A loyal customer base that buys each new version out of habit will not tell the company where its product is actually good — and at some point, habits run out.

The counterintuitive angle sits here: the vagueness of FRITZ 20's positioning may not be a marketing error but a rational short-term decision. When you already have a legacy customer base and an existing distribution channel, releasing a new version to that entire base — from beginner to professional — is the fastest and least risky way to collect revenue. The problem is the long-term price. The broader the audience and the bigger the promise, the more visible the gap between promise and evidence becomes, and that gap erodes credibility faster than any free rival can.

One further point is rarely mentioned: free solutions are not actually free for the learner. Stockfish gives you answers but no curriculum. Lichess gives you an analysis board but no examiner. That gap is the real market for a product like FRITZ 20, and it is precisely where the absence of specifications becomes most regrettable. Data does not lie; the people who read data lie.

I no longer believe in miracles on the board; I only believe in the conversion rate of chances. What I want to see at the next release is not a bigger slogan but a small table: one hundred users, three months, ACPL before and after. A table like that would not reveal any algorithm or give away any secret; it would simply show whether the product does what it claims. If a chess training product is confident enough to call itself a revolution, is it confident enough to publish its own test results?