Carlsen beats Anand on time: the clock decided a drawn position

Short answer

Viswanathan Anand lost to Magnus Carlsen on the icon board of day three of the 2026 Global Chess League without being mated and without dropping a piece: his flag fell on move 64 in a position Stockfish 19 evaluates between −0.14 and 0.00. Eleven moves earlier Anand had 39 seconds to Carlsen's 5. This piece reads the clock move by move from the PGN, measures the positions on the engine, and shows why a 2-second increment makes average time per move — not time remaining — the number that decides.

Event: Tech Mahindra Global Chess League 2026

How this was compiled: Game from round nine, played on 7 September 2026 in Bengaluru. Clock times read move by move from the official broadcast PGN; evaluations measured by us with Stockfish 19 at depth 24, and depth 26 for the move 50 positions.

The essentials in a few lines

• Anand lost on time on move 64 with the position at between −0.14 and 0.00 by Stockfish 19 at depth 26: there was nothing to defend, only moves to deliver.
• From move 41, when the 2-second increment starts, Anand spent an average of 5.9 seconds per move and Carlsen 2.9; from move 50 onwards, 5.8 against 2.1.
• Between moves 53 and 63 Carlsen's clock rose from 5 to 7 seconds while Anand's fell from 39 to 3: under an increment, spending less than it accumulates time.
• Carlsen played 15 moves with 10 seconds or less and 9 with 5 or less, and never flagged; Anand only entered that band on move 58 and flagged six moves later.
• The only serious mistake of the game was 50.g4, which goes from −0.48 to −5.60 after 50...g5; Carlsen replied 50...Bb4 with 8 seconds on the clock and gave it back at −0.43.

What happened: the flag fell in a drawn position

The pairing carried the obvious weight of history: Anand, world champion in 2000 and again from 2007 to 2013, against the player who took the title from him in 2013 and is still world number one. What the board produced, however, was no demonstration. It was a Sicilian with 3.Bc4 and 5.Re1 in which neither side left a half-pawn band until move 27, and an endgame in which Carlsen manoeuvred for more than thirty moves without converting anything.

The evaluation tells the story better than the result does. After 28.a4 the position tilts slightly towards Black — between −0.3 and −0.8 by our measurement — and it is that margin Carlsen tries to squeeze until move 49. From move 53 onwards the engine gives nothing to either side: the reading at depth 24 stays between −0.26 and 0.00 to the end.

The game ended because Anand's clock reached zero after 64.Kh4. There was no mutual time scramble in a murky position, no trap, no hanging piece. There was a player trying to play well a position that had nothing left to play, at a rate the clock could not fund.

The clock as a budget: the increment sets what you may spend

The numbers come from the broadcast PGN, which records the clock after every move. From move 41 to the end Anand spent an average of 5.9 seconds per move and Carlsen 2.9. Restricted to the closing stretch, from move 50, the gap widens: 5.8 seconds for Anand against 2.1 for Carlsen — precisely the value of the increment, which is to say a rate that pays for itself.

The effect is stark in the raw clock readings. On move 53 Anand had 39 seconds and Carlsen 5. Ten moves later Anand had 3 and Carlsen had 7. The player with a second to spare had climbed from 5 to 7 seconds without doing anything special, while the other burned through a cushion that looked comfortable. Anand's single largest expenditure in the stretch was 18 seconds on move 56, in a position the engine already scored at zero.

It is worth stating what this piece does not claim: spending 18 seconds on a move is not a mistake in itself, and Anand's decision to keep playing for a win with level material is legitimate in an event where the match margin counts for tie-breaks. What the numbers show is the price of that choice in this particular format. With a 2-second increment and nothing added before move 41, the budget is the increment, and it is small.

Move 50: both players erred, and the engine measures by how much

The diagram position is bishop and knight against two bishops, pawns on both wings, and the black king already on d5. Anand had to decide how to hold the kingside. The engine's answer is 50.h4, advancing the pawn the dark-squared bishop cannot reach; 50.g4 allows 50...g5!, which freezes h3 and g4 on the colour of Black's bishop and turns the ending into a technically lost one.

Carlsen did not play 50...g5. He played 50...Bb4 with 8 seconds on the clock, and the evaluation snapped back to −0.43. The detail matters because it dismantles the lazy reading of the game: this was not a display of composure under pressure against a rattled opponent. Both players erred on the same move, one by missing the best try and the other by missing the punishment, and neither had the time to look.

The Lichess broadcast's automatic annotation flags 50.g4 as a blunder at −3.11 and 50...Bb4 as a blunder at 0.00. Our own measurement, with Stockfish 19 at depth 26, gives different numbers: −4.50 for the position after 50.g4 and −5.60 for the correct line with 50...g5. The order of magnitude is the same and the conclusion does not change, but the values published here are the ones we measured, not the ones the broadcast displayed.

Twelve dead-drawn moves, with the clock running

From 53.Nf4 the game settles into a pattern of repeated manoeuvres: the white knight tours f4, h5, g7 and e6, Black's light-squared bishop walks d7, e8 and g6, and a pawn trade on the kingside simplifies what little was left. None of it moves the evaluation, and both players knew it — which is exactly why Carlsen was content to play on four seconds.

The contrast in behaviour across this stretch is what the game teaches. Carlsen started playing inside the increment, producing safe moves that worsened nothing, and his clock stopped falling. Anand went on assessing each move as though a choice were still on offer, and his clock swallowed 39 seconds in eleven moves. The player ahead on the clock at move 53 lost on the clock at move 64.

This is the half of the problem that move-quality analysis cannot see: every one of Anand's moves in the closing stretch is fine. The mistake is in no single move, it is in the rate. We measured the other half in this same league, across the 6,199 clocked moves of the first ten rounds: the frequency of serious errors rises from 0.4% above five minutes to 4.1% below thirty seconds. There are two ways to lose when the clock bites, and only one of them shows up on a list of mistakes.

The second flag of the same match

Vidit's game has the same shape plus one extra detail: he reached move 41 with 8 seconds against Erdogmus's 83, and the increment was not enough to recover. The final position, measured by us at depth 24, gives +0.35 to White — a minimal edge, the size of a structural plus, in a middlegame of queens and knights with plenty of play left.

It is tempting to conclude that the two flags decided the match, and the arithmetic is worth checking before saying so. Had both games been drawn, the score would have been 8-5 to the PBG Alaskan Knights rather than 9-7. The winner would be the same; the margin larger. In other words, the flag falls did not change the match result, and Anand's actually cost his own team points.

The margin is no footnote in this league, though: the Global Chess League table is ordered by match points and broken by game points, and the difference between 9-7 and 8-5 can decide who reaches the final. It is the kind of consequence that makes players press on in drawn positions — and that, at a 2-second increment, charges dearly for it.

What to do when the clock gets short

The first habit is about reading positions, not about speed. A position in which any reasonable move holds the evaluation is a position where thinking does not pay: every second spent there is taken from a future moment when there really will be a choice. That was the behavioural difference between the two players from move 53, and it has nothing to do with playing strength.

The second is arithmetic, and it works as an in-game alarm: at a 2-second increment, spending 6 seconds a move costs 4 of cushion per move, and that is what a 39-second cushion funds for ten moves. Knowing that number before the scramble changes when you stop calculating. At a 3- or 5-second increment the arithmetic changes with it — which is why the same player drowns in 3+2 and not in 3+5.

The third is what Anand's game shows by its absence. Trading into a known ending is a decision made with time on the clock, not without it. Once the clock bites, simplification becomes luck: you need the position to cooperate. If your losing pattern is dropping level positions on time, the place to intervene is twenty moves earlier, while there was still a choice.

• Compare your spend per move against the increment, not against the total: under a 2-second increment, spending 6 loses 4 a move.
• When any reasonable move holds the evaluation, play fast on purpose — that is how a cushion is rebuilt under an increment.
• Mark the moves you played under 10 seconds in your own rapid games: if they start before move 40, the problem is in the opening, not the ending.
• Decide the simplification while you can still calculate it; after the scramble it stops being a choice.

Frequently asked questions

Update log

• 2026-09-08 — First publication, with clock times read from the round nine PGN and evaluations measured at depth 24.

Sources

• PBG Alaskan Knights - Alpine APL Pipers: round nine games, with clock times and evaluations — Lichess
• Global Chess League 2026 Day 3: Carlsen's Alpine Pipers Storms Back To Grab Sole Lead — Chess.com
• GCL 2026, Day 3: Alpine APL Pipers seize the lead — FIDE

Related guides

• How to avoid time trouble in chess: five rules for the last three minutes
• Rapid, blitz and bullet: how to adapt your training to each time control
• How to calculate candidate moves in chess: checks, captures and threats before you commit