Blog · August 29, 2026 · Updated September 4, 2026 · 7 min read

Champion scaling in pro LoL: which champions win early and which win late

Open a full draft in Analyze and the card starts with a graph. One line runs from minute 23 to minute 45 and shows which five champions win more at each game length. This post explains where that line comes from, in plain words. The figures are drawn from the same model the card uses, and you can play with them as you read.

What scaling means, as a number

Players say a champion scales when it gets stronger as the game goes on. We turn that idea into two numbers for every champion, fitted on every professional game of the 2026 season.

The first number is the champion's level: how often teams win when they pick it. The second is its slope: how much that win rate moves with every extra ten minutes on the clock. A rising slope means the champion wins more in long games. A falling slope means it wins more in short ones. The slope is what people mean by scaling.

Ornn is the clearest case in this year's data. His teams win about 48% of their games, a little under half. That one number hides what he is good at. In a 25 minute game his win rate is about 43%. In a 40 minute game it is about 57%. He gives up the early game to own the late game, and the slope says so.

Orianna is the mirror image. She wins 52% overall, about 57% in a 25 minute game and about 46% in a 40 minute game. Her edge drains away as the clock runs.

Try Wukong against Bard, or two champions you think of as alike. Most pairs come out close to flat. For most champions, game length changes little.

Which champions scale the most in 2026

Out of the 133 champions with enough games to fit, only ten have a slope we are sure about. Three climb with the clock and three fall. Here they are, read at the card's two clocks: a 25 minute game and a 40 minute game.

Champion	25 min		40 min	Games
Ornn	42.8%	↗	56.9%	660
Milio	44.8%	↗	53.7%	1,127
Hwei	44.2%	↗	51.7%	385
Yuumi	45.8%	↗	51.5%	158
Lissandra	46.6%	↗	52.2%	152
Ezreal	52.1%	↘	46.1%	2,527
Gnar	56.4%	↘	49.4%	1,747
Jayce	51.8%	↘	44.5%	1,390
Bard	56.2%	↘	48.4%	1,757
Orianna	55.3%	↘	46.9%	1,779

Everyone else sits close to flat, or has too few games for us to tell a real slope from noise. That is most of the roster, and it is why the card says "even drafts" far more often than it names a side.

How ten picks become one graph

A draft is two sets of five. We average each side's five curves, then subtract the red side's from the blue side's. That difference is the line on the card. Above the middle, the blue side's picks win more at that game length. Below it, the red side's do.

The band around the line shows how sure we are. Where the whole band is clear of the middle, the colour is strong: that side is favoured at those minutes. Where only most of the band is clear, the colour is faint: that side leans ahead. Where the band straddles the middle, the graph goes gray. Too close to call. A sliver of colour too thin to be worth a word is not drawn at all.

Here is the card's own Draft section for a real game. Team WE drafted Ornn, the biggest scaler in the data, with Azir, Corki and Braum, three more picks that grow with time. Anyone's Legend answered with Orianna, whose edge fades with the clock, and four picks that barely move. Hover the graph to read the gap at any minute.

Read the verdict line first. It says the graph back in words, so the two can never disagree: it names the side the colour belongs to, and where on the clock that colour sits. "Favors" means the whole band clears the middle. "Leans" means most of it does. "Early" or "late" is which end of the graph is coloured, and a line with neither word means the colour runs the whole way. When the two sides swap, the line names both. "Even drafts" means no colour is drawn at all. The two small tables beside it show every pick's win rate in a 25 minute game and a 40 minute game, so you can see which champions do the pulling. Hover a portrait for that champion's game count.

Why ten picks are not ten pieces of evidence

Champions are never picked alone. Teams draft them together and against each other, so the same pairings show up again and again. If we treated the ten picks as ten independent samples, the band would come out too narrow and the card would sound more sure than it should. So we widen the band on purpose. That is also why the wording stops at favors, leans and even. The card never quotes odds.

Why small samples get pulled toward the middle

Say a champion won 12 of its 20 games this season. Is it a 60% champion? Probably not. A fair coin lands heads 12 or more times in 20 flips about one time in four. Twenty games cannot separate a good champion from a lucky one.

So every champion's numbers get pulled toward the average, and pulled harder the fewer games it has. Statisticians call this shrinkage. A champion with fewer than 30 games gets no fit at all, and the card tells you so when you hover its portrait.

Shrinkage is what makes the printed numbers worth reading. We grade the model on games it never saw while fitting, and a printed gap is worth close to what it claims. Without shrinkage the same method prints roughly double the real effect.

Why we fit on every league and every patch

The card fits every champion on every professional game of 2026, and says "all leagues" next to the graph. A fair question: if you are watching the LCK, why is LEC data in your numbers?

We tested that in August 2026. We refit the model on the tier 1 leagues only, on single leagues, and on the last three patches. Then we graded every version on games it had not seen. Every narrower model did worse, including on the very games it was built for.

A single league is only a few hundred games. With that little data the model could not tell one champion's scaling from another's, so it gave every champion the same slope, and its calls were no better than a coin flip. Cutting down to the last three patches was worse than throwing away the same number of games at random. There was no league-specific signal to find, and more games beat fewer every time.

What the graph cannot tell you

It cannot tell you who will win. Team strength predicts a pro game far better than the draft does, and the graph only describes how these ten champions' games went this season. The verdict says so right under it: this does not take team or player strength into account.

Take the game above. Anyone's Legend were favored on the draft, the clock was against them, and they won in 26 minutes. That fits the graph, but the graph did not predict it, and plenty of games go the other way. When the card says even, believe it. The picks do not decide that game.

Try it yourself

A few things to do with this on the site:

• Open Analyze, enter ten picks and read the graph. While a pro game is on the air, click it and the form fills itself with the live draft.
• Hover any portrait in the pick tables to see that champion's game count and its win rate at both clocks.
• Press Share on a section to copy a picture of it. The picture is drawn in your own browser, with no link and no upload.
• Ask a question in plain English, such as "What is Ornn's win rate in the LPL in 2026?", and get the exact number, computed from the games, with those games behind it as a receipt.
• Every champion has a page. Start with Ornn and Orianna, or browse all of them.

Questions about the method are welcome on the contact page.