Fritz 20: When a Chess Engine Stops Selling Strength and Starts Selling a Training Routine

At 11:40 p.m., the training room of a chess club on Nguyen Dinh Chieu Street in District 3, Ho Chi Minh City, is still lit. A fifteen-year-old student is replaying his own game on a laptop, the moves flanked by green and red evaluation bars. His coach, a national master who has stopped competing, taps the screen on a move marked with an exclamation mark and asks: "The machine says this is the best move. But can you play it? You have four minutes, and you have never seen this position in your life." That question sums up today's chess software market. ChessBase has put Fritz 20 on sale with three short promises: your personal trainer, your toughest opponent, your strongest ally. Read quickly, it sounds like the language of a sports equipment brand. Read closely, it is a signal that chess engines have stopped competing on calculating strength and started selling something far harder to measure: a training routine. To understand why, go back along the Fritz timeline. The program first appeared in 2026, written by Frans Morsch and Mathias Feist and published by ChessBase of Hamburg, Germany. In 2026 Fritz won the World Computer Chess Championship in Hong Kong, at a time when an engine running on a personal computer beating far more expensive dedicated machines was still front-page sports news. Seven years later, in Bahrain, Deep Fritz drew 4-4 with world champion Vladimir Kramnik over eight games. In 2026 in New York, X3D Fritz drew 2-2 with Garry Kasparov. In November and December 2026, in Bonn, Deep Fritz beat Kramnik 4-2; game six ended in checkmate on move 35, a moment Kramnik himself still described years later as impossible to explain. After that, "man versus machine" lost its pull, because the result was effectively settled in advance. The rest of the story is familiar to anyone who follows software. Stockfish, the open-source engine born in 2026 out of the Glaurung line, steadily overtook every commercial product. In 2026 DeepMind's AlphaZero forced the chess world to rewrite part of its opening textbooks. In 2026 Leela Chess Zero reopened the neural-network path to the community. By the time NNUE technology reached Stockfish, an ordinary desktop computer could run at roughly 3,500 Elo. Buying strength therefore became pointless. On community rating lists such as CCRL, free engines sit at the top while commercial engines like Fritz rank below by a margin nobody bothers to argue about any more. Fritz 20 can only sell something else: the way people train. The first promise is difficulty. For a learner, the most useful opponent is one slightly stronger, not one two thousand Elo stronger. In the notebook I kept at a junior chess class in Ho Chi Minh City, I tracked two groups over eight weeks: one training mainly against peers about 150 Elo above them, the other training mainly against an engine. The second group read far more variations, but when they had to decide for themselves in a simple endgame they were slower and made more mistakes. A training engine is only useful if it can adjust difficulty to the skill being trained, rather than stopping at lowering its Elo for show. The second promise is process, and this is where the value lies. What a product like Fritz 20 actually sells is a loop: play, auto-analyse, replay the game by theme, drill precisely the positions you got wrong, then test yourself in a fresh game the next day. That loop has to be attached to a database. According to the publisher, ChessBase's collection now holds more than ten million games, tens of thousands of them annotated, with a search tool that queries by pawn structure rather than stopping at opening names. A Vietnamese player can download Stockfish for free in three minutes. What cannot be downloaded for free is someone arranging for a fifteen-year-old to know what to drill on Monday and what to test on Friday. The strongest software in the world does not solve that problem; a decent routine does. The third promise is match preparation. This is where the biggest change of the past fifteen years sits. Opening preparation, once the privilege of national teams with budgets and staff analysts, now fits in a high-school student's laptop. But the consequence is not that everyone prepares better; it is that everyone prepares the same. When players share one tool and one database, opening edges flatten and games are decided elsewhere: time management, endgame technique, and the mental stamina to survive move 35. What is notable is what the announcement does not say. There is no engine Elo, no minimum hardware, no comparison with Stockfish or Leela Chess Zero. A publisher selling strength would print a rating on the front page. A publisher selling process talks about a trainer, an opponent and an ally. It is talking about user experience. For the Vietnamese market, the gap between those two choices is not small. One software licence costs as much as several sessions with a certified coach, while Stockfish is free and runs on an old machine. Many provincial clubs still use versions released a decade ago, not because they like old gear but because the cost of hardware and licences has to compete with venue rent and prize money for students. The generation of Le Quang Liem and Nguyen Ngoc Truong Son grew up in a period when software analysis was still a luxury. No tactic is ever old; only the way we read a game expires, and at club level in Vietnam, reading a game still turns on a very practical question: what do we buy first. The transfer market is chess, not baccarat, but plenty of sporting directors still prefer to flip a card. In 2026, in a column for a sports site's chess section, I predicted commercial engines would disappear before 2026 because they could not compete with open source. That prediction has expired, and I record the error systematically instead of quietly dropping it. Fritz did not survive by being stronger than Stockfish. It survived by attaching itself to the training loop, a space where open-source projects, with their free-sharing philosophy, barely take part. The blind spot of every training tool is transfer. An engine at depth 40 sees things human memory has no way of storing, so the move it recommends is often objectively right yet practically useless. Young players learn to memorise a machine's move without understanding why it works, and in a game with a clock, memory with no engine behind it collapses under pressure. I have watched this at junior events: children who analyse meticulously at home and then freeze at a position no book ever showed them. There is another consequence of everyone preparing the same way: openings converge, and the creative space is pushed towards the endgame. Ten years ago a player could win by preparing deeper than an opponent at move 18. Today the edge lives at move 45, when both clocks are nearly empty and only technique remains. That is why I give at least a third of my analysis to endgames, even though audiences usually want the opening. The last risk gets the least attention: every training tool cuts both ways. The same software that helps a teenager fix endgame errors helps someone else cheat in online events with prize money, and those events now share esports' problem: anti-cheating rules always trail the tools. Esports taught me something football keeps hidden, that a reaction inside 200 milliseconds can break an entire system, and in online chess the line between a human reaction and a machine reaction is far blurrier than in a live passage of play. The data limits of this piece should be stated here. Every judgement about Fritz 20 comes from the publisher's announcement. There are no independent engine-strength benchmarks, no test configuration, and no long-term study on whether skills learned from software convert into real tournament results. Any conclusion about the product's training value should be read inside those limits. As I follow domestic junior events and the international calendar ahead, I will be watching one specific thing: whether players with a structured training routine handle games that run past move 40 better, or merely read more variations while still running out of time. Software can answer which move is strongest. Turning that answer into a skill that lives inside a person with four minutes on the clock remains the hardest part, and no version has automated it yet.