Gukesh's Rise in Numbers
Looking at the changes in Gukesh's play since the time he became a GM
After taking a closer look at Ding's games, I also wanted to look at the games of his challenger Gukesh. For Ding I mainly tried to find out what lead to his poor performance in the last year. But the situation for Gukesh is very different. He is in great form and so young that it's difficult to compare his games from today to them from even one year ago.
So I thought that I'll look at his rise from one of the youngest grandmasters in history to becoming the challenger for the world championship.
Rating
Firstly I wanted to see how Gukesh's rating developed since 2019 (the year an which he became a grandmaster) and how it compares to his performance rating. I decided to calculate the performance rating for each quarter to have a few games to work with while also updating it frequently enough to see potential improvements in his play. I found the results very interesting:
His performance rating was in most quarters higher than his actual rating, which would suggest that his rating hadn't caught up to his actual strength. Especially in 2024 his performance rating is extremely strong.
There are some quarters where he performed below his rating but this has to be expected. It's interesting to see that even these dips in performance rating are over 2650 in recent times.
I'm not sure if it's a coincidence, but for the last 3 years, Gukesh performed worse in the 4th quarter of the year. Note however that for 2024 he only played in the European Club Cub in the last quarter so there are fewer games than in previous years and the data should be taken with a grain of salt.
Quality of play
The logical follow-up question is how the quality of his play changed as his rating went up. But here the analysis gets a bit more difficult.
Since Gukesh improved so much, he naturally started playing against stronger opposition. The average rating of his opponents in the years 2019 to 2024 were 2435, 2376, 2415, 2503, 2645 and 2714 respectively. So as Gueksh improved, his opponents got stronger as well.
This makes the analysis of the quality of his play much more difficult since playing accurately against 2700 opponents is much more difficult than playing with the same accuracy against 2400s. I still decided to look at his average accuracy for each year.
There is a significant improvement from 2019 to 2020 but after that the accuracy stays roughly the same. As mentioned before, playing accurately against stronger opposition is more difficult so that his accuracy didn't decrease as his opponents got stronger shows that Gukesh's play improved.
It's also interesting to see that his accuracy was highest in 2022. This was the year where he had his first very strong olympiad and looking at the rating chart from above, this is also the year he arrived in the top 20 rating wise. 2024 is another strong year for Gukesh which bodes well for the upcoming match.
I also wanted to see how the relative number of his inaccuracies, mistakes and blunders changed over the years.
The relative number of inaccuracies, mistakes and blunders has come down quite a bit since 2019. Other than that, I find it a bit difficult to draw conclusions from the fact that he seems to make more blunders than mistakes. Maybe this stems from the fact that engines recently started having much more extreme evaluations.
Conversion and resilience
Ever since looking at the last candidates tournament, my favourite stats to look at are how many good positions players convert and how many difficult positions they hold.
Again, this comparison is a bit difficult for Gukesh since his opponents have improved a lot over the years, but I still want to take a look at it.
This plot shows that Gukesh is getting less good positions in the last two years and his conversion percentage has gone down a bit. As I said, this has to be expected as his opponents in 2022 had an average rating of 2503 compared to 2714 in 2024. That the number of good games has actually gone up from 2023 to 2024 shows how strong Gukesh has been playing recently.
The numbers of worse and lost games are a bit less predictable.
In contrast to the previous graphs, the number of lost and worse games show no trend over time. But it’s again clear that 2024 has been an amazing year for Gukesh, even though he faced the toughest competition.
Conclusion
I found it fascinating to look at Gukesh’s rise since becoming a grandmaster. One thing that seems clear is that he made another big jump this year which is a good sign for his chances in the world championship match.
In the future I’d like to compare different strong players while they were coming up with each other to see how they would have compared in younger years. It’ll certainly be interesting to see how the rise of the new generation compares to players before them.
Your page is such a great find just before the match. Thank you! Would love to see some more data dives on both players.
not done reading. but something popped up about quality of play (assuming some engine notion of accuracy as in lichess, error bins are based, that there is only one reference level best move by engine) depending on the opponent strength as measure by their cumulative rating estimate. While without such accuracy measure we would naturally agree with that, this might mean something about same band rating notion of quality of play. Is it about the oppoennet serving more accurate positions that demand a more accurate reply, which is possible. My questin is whether that does not point at something about quality of play having some relative aspect to it. I am thinkin about learning theory that could be adjustable and have refined measures instead of the saturated amplitude that comes from referring always to the ceiling. I might have been narrow minded in interpreting "quality" of play here. this was just an idea.