New rating
How the Total Chess Rating works
The Total Chess Rating is one number for how strong a player is across all three formats of the tour: classical, rapid and blitz.
Kjell Madland, Norway ChessMehmet Mars Seven, Norway Chess and King's College London
Draft for public feedback · 1 October 2026
In short
- A chess rating is a number for how strong a player is. Higher is stronger, and the best players in the world are rated above 2700.
- Chess is played at different speeds: classical games last hours, rapid games under an hour, and blitz games a few minutes. FIDE, the world chess federation, gives every player a separate rating for each.
- The Total Chess Rating combines the three into one number, so you can see at a glance who is strongest across all three. A player rated 2700 in all three has a Total Chess Rating of 2700.
- The gap between two players' ratings tells you what to expect when they meet. With a 100-point lead, a player is expected to score about 64 points from 100 games, counting a draw as half a point.
- The new system is separate from and does not affect players’ official FIDE ratings. FIDE will observe the Total Chess Rating during the pilot period, following discussions within the Working Group between Norway Chess and FIDE.
It was made by Norway Chess and its founder Kjell Madland together with game theorist Dr Mehmet Mars Seven of King's College London.
The formula
Let a player have classical, rapid, and blitz ratings , , and , respectively. We define the player's Total Chess Rating by
This formula gives a single number that combines a player's strength across the three formats. If the Total Chess Rating is required to satisfy some natural properties of the Elo system, the formula above is uniquely determined.
Try it
Each rating stands for a playing strength, . The Total Chess Rating is the rating whose strength is the average of the three. The dashed line marks that average.
Total Chess Rating
2830
Plain average of the three: 2828
An example
As an example, consider the classical, rapid, and blitz ratings of Carlsen (2823, 2801, 2860) and Gukesh (2710, 2686, 2659). Using the formula above, their Total Chess Ratings are 2830 and 2686, respectively. The usual Elo formula gives Carlsen an expected score of
Thus, over a 100-game multi-format match, Carlsen would be expected to score about 70 points. This closely approximates the expected score obtained by selecting one of the three formats uniformly at random and applying the usual Elo formula for that format, which in this example is approximately 0.693.
Interpretation
The Total Chess Rating should be interpreted as the single Elo rating corresponding to a player's combined strength across classical, rapid, and blitz. In other words, it is a one-number summary of a player's overall multi-format strength. It is designed for comparing players across formats.
The Total Chess Rating answers the following question: If we want a single Elo rating that summarizes how strong this player is overall across classical, rapid, and blitz, what should that rating be? It has four main features:
- It assigns each player a single scalar rating;
- It yields standard Elo predictions from rating differences;
- It is nonlinear in the original classical, rapid, and blitz ratings;
- It can approximate the expected score obtained by uniformly averaging the three format-specific Elo expected scores.
Why this formula?
As mentioned, the Total Chess Rating formula is uniquely determined if we require the combined rating system to satisfy the natural properties of the Elo rating system. The division by 3 is simply a normalization that keeps the result on the same Elo scale as the inputs: if all three ratings are equal, then the Total Chess Rating equals that rating.
Top 10 players
| Rank | Name | Classical | Rapid | Blitz | Total Chess |
|---|---|---|---|---|---|
| 1 | Magnus Carlsen | 2823 | 2801 | 2860 | 2830 |
| 2 | Hikaru Nakamura | 2792 | 2738 | 2800 | 2779 |
| 3 | Nodirbek Abdusattorov | 2780 | 2704 | 2819 | 2774 |
| 4 | Alireza Firouzja | 2757 | 2737 | 2797 | 2765 |
| 5 | Javokhir Sindarov | 2782 | 2747 | 2725 | 2753 |
| 6 | Arjun Erigaisi | 2746 | 2740 | 2766 | 2751 |
| 7 | Ian Nepomniachtchi | 2719 | 2748 | 2765 | 2745 |
| 8 | Wesley So | 2770 | 2699 | 2745 | 2740 |
| 9 | Fabiano Caruana | 2784 | 2682 | 2738 | 2740 |
| 10 | Praggnanandhaa R | 2763 | 2730 | 2712 | 2736 |
Table 1 reports the Total Chess ranking together with each player's classical, rapid, and blitz ratings. The full list is the Total Chess Rating top 100.
A comparison with the random-format system
Let and denote the classical, rapid, and blitz ratings of two players. Suppose one of the three formats is selected uniformly at random and the game is then evaluated using the usual Elo formula for that format. The resulting expected score is
This differs from the expected score induced by the Total Chess Rating:
The two results are close when the format ratings do not differ too much, as in the Carlsen-Gukesh example. However, they can be different: under uniform random-format selection, the profiles
produce a cycle in the sense that is favored over , is favored over , and is favored over , each with an expected score of 0.609. No scalar rating system can represent such a cycle.
By contrast, all three players have the same Total Chess Rating, which predicts an expected score of 0.5 in every pairing.
Random format
Total Chess Rating
Appendix
Characterization of the Total Chess Rating
This appendix gives a formal justification of the Total Chess Rating formula used in the main text. Recall that the ordinary Elo expected score of a rating against a rating is . Equivalently, if , then . Thus the corresponding Elo odds are .
For each integer , let be a rule assigning one “combined rating” to an -tuple of Elo ratings. For the three chess formats in the main text, the relevant case is . We impose the following properties.
Properties
1. Same-scale normalization
If all ratings are equal, then the combined rating equals that common rating: for every .
2. Regularity and monotonicity
For each , the function is continuously differentiable and has strictly positive partial derivatives: for every and every .
3. Aggregation consistency
The final combined rating is independent of how the ratings are grouped during aggregation, provided that each block aggregate is counted with the multiplicity of the block it represents.
Formally, let be an ordered partition of into nonempty blocks. Here “ordered” means that the blocks themselves are listed in the order . Within each block, however, we use the increasing-index order. Thus, if , we write . Define . Then , where means that is repeated times. For example, .
For another example, suppose four ratings are grouped into two pairs. Let and . Aggregation consistency gives
Moreover, let . Applying aggregation consistency to using the ordered partition , and then applying same-scale normalization, gives
Thus the two equally sized block ratings are themselves aggregated with equal weight.
Aggregating in stages must therefore never change the final combined rating. Intuitively, this rules out hidden dependence on procedural choices: the result should depend only on the original ratings, not on the order in which they are combined.
4. Marginal Elo-strength consistency
For two ratings, the ratio of marginal contributions to the combined rating equals the corresponding Elo odds: .
Intuitively, this reflects the fact that Elo ratings are logarithmic rather than linear. A fixed increase at a high rating level represents a larger increase in underlying Elo strength than the same numerical increase at a lower rating level. For example, holding the other ratings fixed, improving one rating from 2800 to 2850 has a larger effect on the Total Chess Rating than improving that rating from 2500 to 2550.
Thus equal ratings have equal marginal influence, while for unequal ratings the ratio of their influences matches the Elo odds between them. This is the property that anchors the combined rating to the Elo system.
Lemma and theorem
Lemma 1 (Level-set representation). Let be continuously differentiable functions. Suppose that is nowhere vanishing, that is everywhere parallel to , and that every nonempty level set of is connected. Then is constant on every level set of . Consequently, there exists a function on the range of such that
Show the proof of Lemma 1
Proof. Because is nowhere vanishing, every nonempty level set of is a continuously differentiable hypersurface. Since each such level set is connected, any two of its points can be joined by a piecewise continuously differentiable path lying in the level set.
Let be such a path. On each continuously differentiable segment, is constant, so
Since is parallel to , it follows that
Hence is constant along the path. Therefore is constant on every nonempty level set of . We may consequently define to be the common value of on the level set
□
Theorem 1 (Characterization). The unique combined rating rule satisfying the four properties above is given, for every , by
Show the proof of Theorem 1
Proof. Write and . For , normalization gives , which is the stated formula. Now consider the binary rule . By marginal Elo-strength consistency, . The function has the same marginal ratio: . Hence the gradients of and are everywhere parallel. Moreover, is nowhere vanishing, since both of its components are strictly positive, and every level set of is connected: under the change of variables and , it becomes the line segment
The Lemma therefore implies that is constant on the level sets of . Thus
for some function on . Normalization gives . Since is a bijection from onto , the entire range of , it follows that
Now fix and write . Aggregation consistency applied to singleton blocks implies permutation invariance. Hence it is enough to compare the first two coordinates. Using aggregation consistency for the partition , we get . Differentiating with respect to and gives
and
Since the common factor is positive, . By permutation invariance, for every pair , .
Let . Since
the previous display shows that is everywhere parallel to . Moreover, is nowhere vanishing. The level sets of are connected because, under the change of variables for , they become the simplices
The Lemma therefore implies that is constant on every level set of . Consequently, there exists a function on , the range of , such that
Normalization gives
Since is a bijection from onto , the entire range of , it follows that
Therefore,
Equivalently,
It remains only to verify that this rule satisfies the axioms. Same-scale normalization is immediate. The rule is continuously differentiable and strictly increasing because . For aggregation consistency, if , then . Thus replacing each block by copies of preserves the overall average of the 's. Hence the final combined rating is unchanged. Finally, for , . Therefore all four axioms hold, and the formula is unique.
□
Tell us what you think
Total Chess Rating is the rating system of the Total Chess World Championship Tour. This draft is released for public feedback. Comments and feedback are welcome at the e-mail address mehmet.mars.seven@kcl.ac.uk.
Read more about Dr Mehmet Mars Seven, or see the Total Chess Rating top 100.

