FRITZ 20 and Vietnamese Chess's Hard Question: When the Strongest Tool Is Already Free

Last June, I sat behind a twelve-year-old boy in a junior chess class in District 3, Saigon. He had just lost a blitz game at the city youth championship. His coach, an old acquaintance of mine from the years when I organised grassroots tournaments at the Children's Cultural House, said exactly one sentence: Turn on the machine and look again. The boy pulled out his phone, opened a website, pasted the moves in. In under twenty seconds the whole game appeared, red arrows circling the mistakes, an evaluation bar running alongside every single move. No installation. No disc. No licence. No money. I turned to the coach: When did you last open Fritz? He thought about it. Three years ago. And only to look up a few old games in the database. That detail is not dramatic, but it contains the entire problem ChessBase faces with FRITZ 20: the most famous chess software in history is selling something most players already carry in their pocket, for free, and in many cases stronger. CONTEXT: FROM A BLITZ MATCH IN HONG KONG TO A BROWSER TAB Fritz was not built to be a textbook. It was built to beat people. The engine's calculation was written by Frans Morsch, its interface by Mathias Feist, and it was published by ChessBase, the software house founded in Hamburg in 2026. According to records published by ChessBase itself, Fritz beat Garry Kasparov 4-2 in a blitz match in Hong Kong in 2026. I flag the source here deliberately, because that figure comes from the publisher, not from an independent third party, and my habit is to mark that kind of data. Seven years later, in 2026 in Bahrain, Deep Fritz drew 4-4 with Vladimir Kramnik. Four years after that, in November and December 2026 in Bonn, Deep Fritz beat Kramnik 4-2. That was an era when a commercial chess program could still make front-page news, because a machine beating the world champion was still strange. What built the Fritz brand across two decades was not purely strength. It was the interface. Fritz was the first program a club player in Vietnam could open, see a familiar board, click a button and immediately understand what to do. Back when I did chess commentary for VTC, there was always a machine running Fritz in the studio for cross-checking lines. The technician told me something I still remember: this program is at its best where ordinary people can actually use it. That is the inheritance FRITZ 20 claims. It is also the trap. According to ChessBase's product announcement, FRITZ 20 is described as a chess engine but positioned as something wider: a training revolution, a personal chess trainer, an efficient, intelligent and individualised training solution. The stated target audience runs from ambitious players and professionals at tournament level all the way to players taking their very first steps. In the entire text, I counted no technical specification. No strength measurement. No price. No release date. No comparison with any rival. No checkable indicator of any kind. I sat with that document for a while, because my job is to read things like that and separate what can be verified from what exists to sell. MARKET STRUCTURE: THE PRICE OF STRENGTH HAS GONE TO ZERO To understand where FRITZ 20 stands, you have to look at the market it is entering, not at the product itself. In December 2026, DeepMind's AlphaZero faced Stockfish 8 over 100 games. The published result: 28 wins, 72 draws, no losses. The striking part was not the score but the fact that AlphaZero learned chess by playing itself, without a database of human games. The premise that strength required a library of grandmaster games was questioned in a single afternoon. Stockfish took the opposite road and went further. It is open source, maintained on GitHub by a large community, free to download, free to inspect, free to improve. In September 2026, Stockfish 12 integrated an NNUE neural network, producing a strength jump the analysis community registered almost immediately. Alongside it came Leela Chess Zero, an open-source project launched in 2026 in the AlphaZero spirit. At the same time, free and near-free platforms covered every need. Lichess offers Stockfish-powered game analysis at no cost. Chess.com put automated game review into its basic tier and built an ecosystem around it, including Chessable's spaced-repetition method. In 2026, Chess.com completed its acquisition of the Play Magnus Group, consolidating a large slice of the training market under one roof. The outcome of that decade is simple: analytical strength has become a commodity. What once sold for tens or hundreds of dollars now costs nothing, and the free version is not weaker. I helped organise grassroots chess tournaments in Saigon back in 2026, when every arbiter's desk needed its own computer, its own installation disc, its own licence. Today a district-level youth tournament runs on phones. The cost of world-class chess analysis, in real terms, has fallen from several hundred dollars to exactly zero. When something becomes free, its seller must sell something else. ChessBase knows this. The phrase training revolution in their announcement is the answer to that question. The problem is that the answer has not been demonstrated. WHAT TRAINING IS, AND HOW YOU MEASURE IT This is the part I want to spend the most time on, because it is the part most writing about chess software skips. Performance analysis has standardised metrics. ACPL, average centipawn loss, measures the average deviation of each move from the engine's best recommendation. It says a great deal about whether a player is genuinely improving after a training programme. Alongside it sit the rate of serious errors under time pressure, opening-repertoire coverage measured in prepared position types, and review intervals in spaced-repetition systems. If FRITZ 20 is genuinely a training revolution, it has to touch at least one of those four, and it has to publish the result. A trial group. A before-and-after comparison over twelve weeks. A chart showing ACPL declining. Anything another person could check. None of that appears. I am not saying this to belittle the product. I am saying it because this is the minimum standard sports data analysis adopted long ago. When a football club announces that a new training method improved its transition play, it must show pressing numbers, ball recoveries, passes into the final third. Nobody accepts a slogan. Chess is no different. An evaluation bar on a screen can lie. Five consecutive misses of the best move in the same position type cannot. If a training tool does not change the frequency of errors in a specific position type, it has not done any work. That is also what my own analytical career taught me. During two years of chess commentary for VTC, I learned that viewers do not need to know what the engine thinks. They need to know what the player thinks, and the engine exists to check that thinking. Software that sits between those two things is, at best, a translator. At worst, it is a results reader. FRITZ 20 claims to be the first. Its announcement only provides evidence of the second. THE PROBLEM WITH A PRODUCT FOR EVERYONE One small detail in the announcement made me pause longer than anything else. This product is for players at tournament level and professionals, and also for players taking their first steps. In software generally, and in chess specifically, those two groups have fundamentally different needs. Beginners need to be steered away from mistakes in the opening, need exercises with clear feedback, need a path that does not overwhelm them. Professionals need tools to dig into position types only a few hundred people on earth have ever seen, need deep customisation, need processing speed and accuracy that stress the machine itself. An interface that delights the first group usually irritates the second, and the reverse. A product claiming to serve both tends to end up stretched: good enough to use, never sharp enough to need. Look at the market and the successful products have all picked a spot. Stockfish owns raw strength, free and open source. Structured learning platforms own the curriculum. Integrated platforms own the overall experience, where analysis, play, study and community live in one place. A product standing between all three, unclear about which one it occupies, will struggle to answer the buyer's simplest question: what am I buying this for, and what does it replace. I once heard a young coach in Saigon say he would happily pay for software if it reduced the number of errors his students make in one specific position type. He did not want a revolution. He wanted a number going down. No tactic is ever old; only the way we read the position expires. The same holds for software: no feature is old, only the way its value is proven has expired. Declaring a product a revolution was effective marketing last decade. At a time when users can measure everything themselves, it becomes a debt of trust. DATA LIMITS I add this section to every analysis, and this time it matters more than usual. The FRITZ 20 announcement I have carries no publication date. That means I cannot determine whether this is a new product still inside a promotional cycle or a long-standing product page maintained for search queries on the product name. Without a time anchor, any conclusion I draw about its relevance is a conditional estimate. The source is the publisher, a party with a direct commercial interest. That does not make the information false, but it forces every product description into the promotional category, requiring independent verification before it can support a purchase decision. There is no independent test, no comparison against free engines, no price, no system requirements. In 2026, when the pandemic stopped world football, I had to discard half my old dataset because post-lockdown football was a different sport. I retell that not to make it a rule for every case, but to separate two kinds of information. The noise to discard was modelling built on a dense fixture calendar that no longer existed. The principle that survived was basic: to evaluate a change you need a before and an after, and you need to remove confounders. In chess, the largest confounder of the past decade is that elite analysis became free. Any assessment of a paid product must account for that. THE COUNTERINTUITIVE ANGLE: THE REAL ADVANTAGE SITS OUTSIDE THE ENGINE There is another reading of this whole story, and I think it is the more accurate one. The conventional reading says ChessBase is losing a strength race. True, but irrelevant, because the strength race is over and the prize went to the free side. The second reading says ChessBase's real asset was never the engine. It is the human-curated game database: millions of games, annotated, classified by opening, tagged with players and events. An engine can read a position in milliseconds, but it cannot tell you which line that game belongs to, who has played it, and where those players went wrong and how. That is human work, and no open-source community does it for free at comparable scale. If FRITZ 20 wins anywhere, it will win at that translation layer: where the engine connects to the database, to competitive history, to the user's learning. The word training in their announcement, in the end, points in that direction. But there is a flip side the announcement never mentions. The same process that made elite analysis free and instant also made cheating far cheaper. A club player today carries a tool stronger than any computer that sat in a tournament hall fifteen years ago. Anti-cheating regulations at international and national level are still chasing that reality, with a lag measured in years. This is the kind of problem I have seen in another field, where the spread of technology outruns the writing of rules. When the cost of doing something wrong falls to near zero, personal ethics become the only line of defence, and that line does not hold under pressure at scale. For chess, it means the real value of a training product is not only whether it makes a player stronger. It is whether it teaches a player to distinguish understanding from copying. A tool that teaches people to copy the machine's moves produces players who are better in the study and weaker at the board. A tool that teaches people why that move is correct produces the opposite. TAKEAWAY: A TEST THAT CAN BE CHECKED I do not intend to conclude whether FRITZ 20 is good or bad. I do not have the data to say that, and I do not want to repeat a mistake I have made before: judging something too early on impressions. What I propose is a testable wager over the next twelve months. If FRITZ 20 is genuinely a turning point in training, the publisher will release a before-and-after measurement on a defined user group: how much average centipawn loss fell, how much the rate of serious errors under time pressure fell, how much opening coverage grew. Numbers, sources, a control group. If that does not appear, then all we have is a claim, and a claim never made anyone stronger. Last June, the twelve-year-old in District 3 will still be sitting there with his phone, knowing within twenty seconds where he went wrong. The chess industry's problem is no longer making the machine stronger. It is making the human understand more than the machine, precisely where the phone cannot speak for them.