Fritz 21 and Powerbook 2026: what the opening book measures, and whose human moves it keeps

Short answer

ChessBase released Fritz 21 in May 2026 and the Fritz Powerbook 2026 in August, and the pairing is being described as a way to study openings against real human replies rather than synthetic engine lines. The description is accurate and the caveat matters: the Powerbook's human moves come from 1.7 million high-class tournament games, and one of the books in the bundle uses only players above 2550. This piece sets out what each product is, what ChessBase's own reviewer records as its limitations, and measures where the three icon-board games of the Global Chess League were actually decided.

How this was compiled: Compiled on 11 September 2026, with prices and descriptions read from ChessBase's product pages and reviews — Fritz 21 released in May 2026, Powerbook 2026 in August 2026 — and with the game figures measured by us using Stockfish 19 at depth 24 with three lines per position, across the three icon-board games of the Global Chess League on 10 and 11 September 2026. The database sizes quoted are the manufacturer's own; we have no way to audit them.

The essentials in a few lines

• The Fritz Powerbook 2026 claims 25 million opening positions drawn from 1.7 million high-class tournament games, storing for each position every move played there, the average rating of the players who played it and the score each one produced.
• The bundle includes an exclusive book of grandmaster games rated 2550 and above from the past hundred years — 2.7 million positions — and a CorrBook drawn from the best 100,000 correspondence games.
• There is no product called "Openingbook 2026": ChessBase's names are Fritz Powerbook 2026, released in August 2026, and Opening Encyclopedia 2026, released in May.
• ChessBase's own reviewer records three limitations in Fritz 21: dependence on the firm's online services, thin grounds for upgrading from Fritz 20, and AI playing styles that "cannot replace a human training partner capable of genuine psychological decisions".
• In the three icon-board games of the Global Chess League that we measured, none was decided in the opening: all three turned between moves 16 and 35, each on a single middlegame or endgame move.

What was released, and under which name

It is worth starting with the name, because it circulates wrongly. There is no "Openingbook 2026". ChessBase has two opening products carrying that year: the Fritz Powerbook 2026, which is a move book with statistics, and the Opening Encyclopedia 2026, which is a theory library — over 1,500 thematic databases, around 8,200 surveys organised by ECO code, more than 40,000 model games and 100 videos totalling 35 hours. They are different things with different uses, and people looking for one often buy the other.

Fritz 21 itself is a playing and training program, not an opening product. What it adds over Fritz 20 sits in the training layer: setting whether the program should play aggressively, positionally, exchange-oriented or with risk; a calculation trainer under time pressure; and natural-language commentary during play. The engine gained around 40 Elo over the previous version, on the manufacturer's figure.

The question that matters to anyone studying is not which bundle holds more gigabytes. It is whose moves the book keeps — and that answer is on the product page, not in the marketing.

What the Powerbook 2026 actually keeps

That difference is real and it is the best thing about the product. An engine answers "what is the best move"; a statistical book answers "what do people play here, at what rating, with what results". The two questions have different answers in most opening positions, and the second is the one that decides games — because the person on the other side of the board is not the engine.

The 2026 bundle carries three books besides the main one, and the labels matter. The Strongbook and the CorrBook filter the sample: the second comes from the best 100,000 correspondence games, which means moves chosen with days of thought and computer assistance — the precise opposite of a human move under a clock. There is also a book the maker calls small and exclusive, made of grandmaster games rated 2550 and above from the past hundred years, totalling 2.7 million positions.

Put the two descriptions together and the picture is clear: the sample is human, but it is human of a specific kind. "High-class tournament games" and "Elo 2550 and above" do not describe the opponent of someone playing at 1500 on Lichess or Chess.com on a Saturday night. That does not invalidate the book — it invalidates the promise that it teaches you to answer what you are going to face.

Who the sample serves, and who it does not

The case is easy to recognise. You study a main line to move 12, sit down at the board, and at move 5 your opponent plays something the book records in 0.3% of games. The question you need answered — what does that move give up, and how do I punish it — is not in the book, because its sample is made of people who do not play it. The book is excellent for knowing what is fashionable at the top and why; that is also what the Opening Encyclopedia delivers, in prose.

Anyone studying alone has two cheaper alternatives better aimed at the problem. The first is to use the game database of their own rating band, which the large sites expose with a rating filter — that is where the move played 0.3% of the time at the top turns up in 8% of games. The second is to start from their own losses: the positions where you have already left the plan are, by definition, the ones you will meet again.

None of this is an argument against buying the bundle. It is an argument against the promise attached to it. A book of 25 million elite positions is a formidable reference instrument, and a poor substitute for knowing what your own opponents do.

What ChessBase's own reviewer wrote

The reviewer's three caveats are specific and worth quoting as they stand. First: "some of the new features reveal their full potential only when used together with ChessBase's online services", which means part of what you bought depends on a subscription and a connection. Second, the upgrade question. Third, and pointed at the product's central promise: "the different playing styles are recognizable, but they cannot replace a human training partner capable of genuine psychological decisions".

That third sentence is the most useful of the three, and it speaks straight to the previous section. The program can imitate the shape of a style — playing for exchanges, taking more risk, preferring structure — but it does not reproduce the reason a human errs: the clock, the expectation that a position will be easy, the refusal to trade when material up.

It is rare for an in-house review to record limitations with that degree of precision, and it deserves credit. Anyone who reads the whole note before buying makes a better decision than someone who reads the headline.

Where this week's games were actually decided

In Carlsen's win over Sindarov, an Arkhangelsk Ruy Lopez, Black's three best continuations at move 14 measured between +0.34 and +0.55; the game turned at move 16, on 16...c5. In Carlsen's win over Anand, a King's Indian Attack, the first measurable advantage was Black's — -0.69 after 21.c4 — and what decided it was 28...Nd4 in a position measuring 0.00. In Sindarov's win over Nepomniachtchi, a Queen's Gambit Declined, White's edge reached +1.43 at move 17, fell to +0.48 by move 21, and only became a won game with 29...Ne4.

The conclusion is not that openings do not matter — these players reached move 14 in a playable position precisely because they know the theory. The conclusion is about where the marginal return on study sits. For a player who no longer loses games in the opening, the next rating point does not come from memorising five more moves of a line: it comes from the move 28 played in 44 seconds with nine minutes on the clock, which is what happened to Anand.

For a player who still does lose games in the opening, the order reverses, and there the problem is not the Powerbook either: it is knowing how to punish the four or five bad moves that actually turn up in their games. Both cases share a diagnosis and a tool — looking at your own games before choosing what to study.

How to use an opening book without becoming its hostage

The common error is to start at the end. The book is bought, a handsome top-level line is chosen, and fifteen moves are memorised that never appear, because the opponent deviates on the sixth. Study time is the scarce resource, and it returns more when it is spent on positions you have already proved you will reach — the ones in your last fifty games.

When the time comes to consult the book, use the whole statistic and not only the frequency column. The Powerbook stores, per position, the average rating of the players who chose each move and the score each one produced; a move played by 2400s scoring 45% and a move played by 2650s scoring 58% say different things about what is sound and what is fashionable.

And keep the evaluation bar in its place. A move the engine measures at -0.15 can be the best practical move of your life if it produces the structure you know how to play; a 0.00 move leading to a technical endgame you have never studied is a trap you set for yourself. Our guide to studying openings with Stockfish deals with exactly that boundary.

Frequently asked questions

Update log

• 2026-09-11 — First publication, with prices and descriptions read from ChessBase's pages and the three Global Chess League games measured by us at depth 24.

Sources

• Review: Fritz 21 — ChessBase
• Fritz 21 & Fritz Powerbook 2026 — product page — ChessBase Shop
• Review: Opening Encyclopedia 2026 — ChessBase
• Tech Mahindra Global Chess League 2026 | Preliminary Stage — games with clocks and evaluations — Lichess

Related guides

• How to study chess openings with an engine without becoming a slave to the evaluation bar
• How to build an opening repertoire: 1.c4 as White and the Sicilian as Black
• Tactical vision or opening theory: where your study hours actually pay off