Fritz 20: The Training Machine and the Redistribution of Power in Chess

A January morning in Delhi. I am sitting in the practice room of a small chess academy in Saket, watching a thirteen-year-old boy replay a game he lost two days earlier. He blundered on move 34 — a pawn dropped on the queenside in a position his coach called “fine”. Teacher and student spent two days finding it. The machine found it in under a second, and flagged four more errors nobody in the room had noticed.

I sat there, sixty years old, after forty-four years on several sides of this industry — player, tournament organiser, reporter, podcast host — and I thought that the story worth writing this season is not a transfer rumour. It is the screen in front of that boy.

Fritz 20 has just been released with a loud claim: an engine that is more than an analysis tool, a training revolution for ambitious players and professionals, for beginners and tournament players alike. I have read sentences like that for thirty years. What interests me is not the promise. It is the structure beneath the promise, and who pays for it.

Mistakes are not frightening. Bending down at two in the morning to pick the tape back up is what makes a forecaster. In chess the tape is the game already played — and this industry has just been handed one more machine to pick it up.

CONTEXT: FROM FRITZ 3 TO FRITZ 20

Fritz was born in the early 1990s, built by Frans Morsch and Mathias Feist and sold by ChessBase, the German software house. It became famous in 2026 in Munich when Fritz 3 beat Garry Kasparov 4-1 — a commercial engine, sold to amateurs, beating the world champion.

The timeline after that is a set of milestones anyone in my trade knows by heart. In 2026 Deep Fritz drew 4-4 with Vladimir Kramnik. In 2026 X3D Fritz drew 2-2 with Kasparov in New York. In 2026 in Bonn, Deep Fritz beat Kramnik 4-2, the last time a reigning world champion stood level with a machine across a long official match.

After 2026, chess split into two worlds. The human world, where the highest rating ever recorded belongs to Magnus Carlsen at roughly 2,882. And the machine world, where Stockfish, open source, is rated above 3,600 by lists such as CCRL. That gap of some 800 points is not decoration. It is the entire economy of modern chess coaching.

In 2026 DeepMind published AlphaZero and shocked the method itself: a machine learning by playing itself, without a human game database. In 2026 Stockfish integrated an NNUE neural network and lifted the analytical standard again. By then everyone serious understood one thing: analytical strength is no longer something you sell. It is free at the door.

That is the trap Fritz 20 has to live inside. When Stockfish is free, when Lichess gives analysis away, when Leela Chess Zero is open source, a commercial engine cannot sell strength. It must sell something else. The launch copy for Fritz 20 says exactly what: training.

There is a very concrete market reason. India is in the middle of an unprecedented chess boom. In December 2026 Gukesh Dommaraju became world champion at eighteen, beating Ding Liren 7.5-6.5 in Singapore, the youngest champion in history. That same year R Praggnanandhaa and Arjun Erigaisi crossed 2,800. Koneru Humpy won the women's world rapid title. A country with hundreds of thousands of children in small academies — each of them a potential customer for training software.

From my years watching tournaments and junior events in India, I will say what few say: that boom did not come from paid software. It came from free tools, from YouTube, from a cheap coaching network in Chennai and Hyderabad. Which is why a commercial training engine raises a political question more than a technical one.

WHAT “TRAINING REVOLUTION” ACTUALLY CONTAINS

The claim breaks into three layers.

First, an adaptive sparring partner — an engine that can play at 1,400, then 1,800, then 2,100, and play like a human at that level rather than a god pretending to be clumsy. This is the most valuable thing in coaching. When I organised events in Vietnam and later in India, every academy's hardest problem was finding a partner of the right strength. Too strong, the student only learns to absorb punishment. Too weak, the student learns bad habits.

Second, error detection and classification. The machine does not merely say “this move is wrong”. It labels severity, and more importantly it aggregates patterns. A student drops a pawn on the queenside on move 34 this game, on move 31 last week, on move 29 last month. That pattern is the lesson. One game teaches little. Forty games with the same error teach everything.

Third, a personalised opening repertoire — not a giant database, but one that fits the player's own style and is updated for specific opponents in a specific event. That was work players once paid analysts to do by hand over weeks.

The core insight: the bottleneck in chess improvement is not analytical power, it is the feedback loop and the discipline of self-assessment.

Let me go slowly, because this is where most articles about chess engines get it wrong. A strong engine does not automatically produce a strong player. If it did, everyone with Stockfish would be a grandmaster. The distance between “the machine showed me the right move” and “I found the right move in a real game, under clock pressure, after four hours at the board” is the entire distance of improvement. The machine can close the first part. The second part belongs to the human and to habit.

That is why I take Fritz 20 more seriously than a normal person takes a piece of software. If it genuinely turns error analysis into a daily, targeted, tracked loop with a history, then it is not selling software. It is selling discipline. And discipline is the most expensive commodity in this trade.

THE REAL COST OF A TRAINING MACHINE

One small Indian academy I sat in had about forty students paying a few thousand rupees a month. Its head coach earned less than a fresh software engineer in Bangalore. A licensed training package, plus a game database, plus a capable computer, plus annual upgrades, can cost several months of that coach's salary.

Here is the paradox I want named properly: free tools built the Indian chess boom. And now, as those children reach the level where serious coaching begins, the next door has a price on it. That is the business model of the entire professional software industry — give it away at the bottom to build habit, charge in the middle to produce results.

I am not condemning it. I am naming it. A training engine is an investment, and every investment has winners and losers.

The clearest winners are players aged twelve to eighteen, in the fastest-learning and most habit-forming years. For them, an adaptive sparring partner is worth far more than a stronger analytical engine. They do not need to know what move 27 should have been. They need a playing partner who does not frighten them.

The clearest losers are mid-tier coaches. Most of their income comes from pointing out errors and fixing openings. When software does that faster, cheaper and more patiently, the coach's value must move upward: psychology, event selection, schedule management, teaching how to lose. Those are things the machine cannot do.

On market structure: chess has no transfer window in the football sense, but it has a real market — analysts, opening coaches, invitation places. A new tool always gives a temporary edge to whoever buys it first. That edge usually lasts twelve to twenty-four months before an equivalent becomes common or free.

If Fritz 20 is what it claims, its temporary edge lies in the training loop, not in raw strength. That is far harder to copy than adding a few hundred Elo.

An empty stadium can kill emotion, but it cannot kill tactics. Training software can make a practice room silent, but it cannot replace the person sitting in it.

THE OPPOSITE VIEW: THREE THINGS A MACHINE CANNOT DO — AND ONE IT CAN RUIN

First, it cannot teach nerve. I once sat in a tournament hall in Chennai and watched a young player with a winning position collapse in the last ten moves because his hands shook. The machine had shown him the right move three weeks earlier. The machine cannot sit in that chair, cannot feel the clock, cannot know that the opponent stood up and walked around and that this frightened him.

Second, it cannot teach you which battles to fight. It can tell you where you erred in a game. It cannot tell you which events to enter, how long to rest, which invitations to refuse. Over forty-four years I have seen more talents destroyed by schedules than by technique.

Third, and this worries me most: it can create addiction to answers. A good training tool must know when to stay silent. It must let the student suffer for ten minutes before revealing the truth. If software shows the right move too fast, too beautifully, too conveniently, it teaches a different skill entirely — the skill of pressing a button. In academies I have visited, I have seen children who analyse brilliantly and compete terribly. They have memorised the machine's assessment without forming their own judgement.

And here is what it can ruin: it can teach a generation to measure itself by someone else's yardstick. Chess is a sport whose deepest pain is the pain of realising your own mistake. If software removes that moment, it has removed half the value of the game.

The question I asked was not ladylike, but the way they dodged it was not manly either. In 2026, at a press conference at the Jawaharlal Nehru Stadium in Delhi, I was the only woman in the room. I asked about leaving an ageing striker too deep with nobody running beyond him. A male colleague smirked. The next match was lost for precisely that reason. I tell this not to boast, but because it taught me a rule that applies to every new tool, Fritz 20 included: what matters is not the right answer, it is the right question. Any machine can supply an answer. No machine knows which question to ask.

If Fritz 20 is truly a training revolution, it must prove it by teaching users to ask better questions, not by answering faster.

TESTABLE PREDICTIONS

One: by the end of 2026, mid-tier Indian academies will use recurring-error data as their formal curriculum instead of packaged opening files. Parents will stop asking “what opening is my child learning” and start asking “what position type does my child keep failing in”.

Two: coach value will shift visibly from technique to psychology and schedule management. Those who only open an engine will lose students. Those who read the student will raise prices.

Three, and I am ready to be wrong: within the next decade a commercial engine will not compete on strength. It will compete on memory. A machine that remembers a player's error history across ten years will beat a machine that is 100 Elo stronger.

Ten years in the middle of arguments, and the fire lit by that 2026 question still burns — not because I like a fight, but because this industry keeps dodging the questions that matter: who gets access, who pays, and what is lost when every answer is already available.

If you are considering buying Fritz 20 for your child, I will not tell you yes or no. I will tell you to do one thing first: ask your child to analyse a lost game for twenty minutes without opening the machine. If they can, the software is a tool. If they cannot, the software is a crutch. Both carry you some distance. Only one carries you to the end.