Fritz 20 and the Chess Training Revolution: When the Machine Agrees to Play Like a Human

On a late November evening, in a small chess café tucked down an alley off Nguyễn Chí Thanh Street in Đà Nẵng, I sat behind a fourteen-year-old boy for forty minutes without saying a word.

In front of him was an old wooden board; beside him, a laptop already open. On move twenty-three he retreated his knight to f8 — a move I would guess nine out of ten amateur players in Vietnam would skip, because it looks like a step backwards. He clicked. The evaluation bar crept forward a fraction, less than half a pawn, then held still. He exhaled, wrote a line in a notebook with a pencil, and underlined it twice.

What held me there was not the move. It was the note-taking: by hand, slowly, while the machine inside the laptop could print three hundred lines of variations in a thousandth of a second. Four years ago, when I began writing about chess in central Vietnam, I assumed ever-stronger analysis engines would wipe out the notebook. I was wrong. The more computing power is pushed into the machine, the more a player needs somewhere to record what he actually understands.

The revolution Fritz 20 is trying to open sits exactly in that gap: not making the machine stronger, but making it willing to stoop to human height.

Fritz first appeared in the early 1990s, written by Frans Morsch and Mathias Feist and published by ChessBase. In 2026, Fritz 3 won the World Computer Chess Championship in Hong Kong. In 2026, Deep Fritz drew 4-4 with Vladimir Kramnik in Bahrain, a match billed as “Brains in Bahrain.” In November 2026, X3D Fritz drew 2-2 with Garry Kasparov in New York. In November 2026, Deep Fritz beat Kramnik 4-2 in Bonn, and the sixth game ended with a mate-in-one that any club player would have seen; Kramnik missed it with minutes left on the clock. That is both the beautiful memory and the scar: the machine beats humans not through anything miraculous, but by never falling asleep on the final move.

Then chess changed. In 2026 it was Kasparov himself who staged “advanced chess” in León, Spain, where players were allowed to consult computers during play. The idea was condemned as anti-sporting. Twenty-seven years later, nobody trains seriously without an engine at their side. The question has shifted from whether to use a machine to how to use one without becoming its shadow.

In Vietnam, that distance has its own colour. Vietnamese chess has milestones beyond dispute: Nguyễn Ngọc Trường Sơn won the world under-10 title in 2026; Lê Quang Liêm won the World Blitz Championship in 2026 in Khanty-Mansiysk, one of the biggest individual titles the country has ever held. But the grassroots movement behind those two names is uneven.

While Chinese chess still lives on the pavements of Saigon in chalk-drawn boards and the murmurs of onlookers, Vietnamese chess exists mainly in three kinds of space: talent classes at children’s centres, private clubs in Hà Nội, Hồ Chí Minh City and Đà Nẵng, and small training rooms that a few grandmasters open to supplement their tournament income.

On software, the domestic market is a paradox. Stockfish — the strongest open-source engine in the world — is free, and Lichess lets you run it in a browser. So when a product like Fritz 20 sells for the price of a few million đồng, buyers are not paying for computing power. They are paying for the interface, the lessons, a pre-arranged path, and the feeling that a teacher is sitting beside them. At average Vietnamese incomes, that purchase still sits on the line between an investment and a luxury.

Within the annual season, Vietnamese chess follows a predictable cycle: the post-Tết period opens the youth tournaments, mid-year brings national qualifiers, and the year closes with team events and international slots. The gaps between those markers are exactly when training software earns its keep — or when it lies idle on a hard drive.

From my own observation of youth events in central Vietnam this season, one signal keeps repeating: teenagers with their own computers improve much faster at first, then plateau right as the children coached in person begin to overtake them. That signal matters more than any Elo figure.

To understand Fritz 20, it helps to say plainly what it is. This is ChessBase’s chess training suite, heir to a Fritz line more than thirty years old, and the new edition puts its weight on individualised training rather than raw playing strength.

The change is in the engine. Fritz 20 runs on a neural-network foundation: instead of adding up material, scanning squares and scoring central control against a hand-written table, it learns from tens of millions of games to produce an evaluation close to human judgement.

The first consequence is small but vital for beginners: your “unorthodox” moves are no longer condemned as serious errors for no good reason. Older engines could label a perfectly playable plan a blunder simply because theory would say otherwise thirty moves later. A neural engine distinguishes “slightly worse” from “a piece down,” and for a learner that distinction is worth more than a hundred Elo points. A tool that concludes wrongly and loudly teaches its user to be delusional.

The second is the strength range. Fritz 20 lets you pick playing levels from roughly 800 Elo to over 3000, and this is where the real engineering lies.

The old way of weakening an engine — capping search depth — produces random, silly, chess-meaningless moves. You learn nothing from an opponent that hangs a rook. Fritz 20 does it differently: at lower levels it draws on a database of real games to choose the moves that players at that level actually play. Its mistakes are real mistakes, the kind you meet at a youth tournament or a practice game. You are training against a human, not against a machine pretending to be stupid.

The third feature, and in my view the most pedagogically valuable, is in-game self-adjustment. The goal is to keep every game tense, always in the zone where you have to think to win. In training, the optimal discomfort threshold sits between two extremes: not the level where you win every game, and not the level where you are crushed, but the level where you lose roughly half. That is a lesson few Vietnamese parents have been told: handing a child an opponent who is far too strong does not build grit, it builds paralysis.

The fourth is openings. Fritz 20 ships with an opening book and lets you build your own repertoire, test your memory and find holes in your own file.

In Vietnam, the opening is where rote learning shows itself most clearly. I have met eleven-year-olds who have memorised the first twenty-five moves of top-level games, play them at speed, then sit bewildered on move twenty-six, because nobody taught them that an opening is a story rather than a poem to be recited with the right accents. A repertoire builder only helps someone who already understands why they chose that system. For everyone else, it just accelerates the rote learning.

The fifth is annotation and assisted analysis. Fritz 20’s automatic annotation produces comments in human language rather than strings of symbols. The assisted-analysis mode — my favourite — locks the engine while it is your turn to think, so you must find the move yourself before seeing the answer. It sounds trivial, but it recreates tournament conditions, where no machine whispers in your ear. After many years in tournament halls, I believe what kills the progress of young Vietnamese players is not a lack of knowledge but a lack of discipline in thinking alone, in silence.

The sixth is exercises: tactics, endgames, clock work, error statistics by motif. The endgame is routinely neglected in clubs, yet it is where class differences between players of the same age show most sharply.

The seventh is the ecosystem. Fritz 20 connects to ChessBase’s databases, the Playchess server and cloud game archives, so a game you lose online can become training material the same evening. That is where the real strength of chess software lives, not in how many million positions per second it calculates.

A comparison clarifies the position. Stockfish is stronger than Fritz 20 in almost every position, and Leela Chess Zero plays in a stylistically different way worth studying. Both, however, are tools for people who already know what they are looking for. Fritz 20 aims at people who do not: it lays out a path, asks questions, scores you and keeps a training history. The difference between an engine and a training suite is the difference between a dictionary and a textbook. The dictionary is thicker, more complete, and utterly useless to someone who cannot spell.

On hardware, the software demands less than many assume: a mid-range laptop with 16GB of RAM and a solid-state drive handles everyday training. On price, a few million đồng per licence makes many Vietnamese buyers hesitate for a long time, especially with a free and stronger engine standing next to it.

Based on my experience following domestic tournaments, Vietnamese buyers of training software fall into two clear groups. The first buys to use: they open the program three or four times a week, take notes, repeat. The second buys for peace of mind: they believe owning the best tool is already half the journey. The second group is larger, and it is also the group that never comes back to buy the next edition.

The first paradox: the most sellable feature of Fritz 20 is precisely where it is weak. The entire chess industry has chased Elo for three decades, to the point where an engine thirty points stronger is advertised as a transformation. Fritz 20 swims against the current: it sells the ability to play badly in a sensible way.

But most buyers do not notice. In the first week they drag the slider to the top level, lose ten games, feel flattened, close the program and return to casual games. A good tool cannot cure a bad habit, and training software cannot manufacture discipline on behalf of the person using it.

The second paradox lies in collective memory. When people recall the computer era in chess, they recall the games machines won: Kramnik broken in Bonn, Kasparov drawing in helplessness in New York. It is dramatic storytelling, and it obscures something more important.

What gets obscured is this: machines taught players moves no coach had imagined, and a fourteen-year-old in Đà Nẵng today reaches knowledge that twenty years ago required a grandmaster to touch. The generation of Lê Quang Liêm and Nguyễn Ngọc Trường Sơn came of age when engine analysis was expensive and slow. The next generation came of age when it was free and instant. If the achievement gap between those generations has not narrowed accordingly, the problem lies elsewhere — not in the software.

The third paradox, and the one I want to state most plainly: the bottleneck for Vietnamese chess has never been the analysis engine. Anyone with a computer can install Stockfish for free, stronger than Fritz 20 in most positions. Yet our number of grandmasters has not grown at the speed of computing power. The bottleneck is a thin tournament system, meagre prize funds, a handful of qualified coaches, and the lifespan of a chess dream inside a middle-income family.

And there is one more layer, belonging to how the market tells stories. Chess software is sold as a promise of results: buy it and you will improve. That is the same logic as sponsorship contracts that keep athletes speaking safely and blandly to protect an image. Markets always prefer a product that promises over a product that tells the truth. The player, meanwhile, still has to answer the old question: am I buying a tool, or buying the feeling of progress?

Back to the café on Nguyễn Chí Thanh Street. The boy finished his note, closed the notebook, shut the laptop, and sat looking at the board for another two minutes before standing up. I did not ask what he was thinking.

Sport does not promise victory; it promises only a heartbeat willing to continue. And in chess, that heartbeat usually happens in silence, in the exact interval between the machine offering an answer and the human agreeing to understand it.

The microphone does not make the storyteller. It is the silence of the stands that signs beneath every line.

Fritz 20 is a good tool, possibly the best training tool a serious player in Vietnam can buy this season. But if there is one thing a neural network cannot simulate, it is the patience of one person waiting for another to understand something on their own. A machine can show a child the right move on move twenty-three. Someone has to sit there, without rushing, so the child believes the time he spent understanding it was time worth living.