The best prospects to watch next season: players, teams and regions

Teams building the most prospects

Organisations with three or more of the top 50 prospects, counting their academy and challenger teams with them. Each name is followed by the player's potential.

• T1 (8): Peyz 95, Cloud 92 (T1 Esports Academy), Painter 92 (T1 Esports Academy), Cypher 92 (T1 Esports Academy), Aether 91 (T1 Esports Academy Rookies), Haetae 89 (T1 Esports Academy), Eclipse 88 (T1 Esports Academy), Jinbeom 87 (T1 Esports Academy Rookies)
• Dplus Kia (6): Smash 98, Siwoo 94, Jaehyuk 92 (Dplus Kia Challengers), Garden 91 (Dplus Kia Challengers), Solid 90 (Dplus Kia Challengers), Lucid 89
• KT Rolster (4): FenRir 94 (KT Rolster Challengers), Pollu 91 (KT Rolster Challengers), Hwichan 89 (KT Rolster Challengers), Sero 89 (KT Rolster Challengers)
• Nongshim RedForce (3): Diable 93, Janus 91 (Nongshim Esports Academy), MihawK 88 (Nongshim Esports Academy)

Most of these players are in academy leagues (LCK Challengers), not on the main roster yet. That is where potential is most useful: their level against academy rivals is adjusted for the league, and the model has learned from earlier players in the same leagues.

Regions producing them

Korea has 18 of its 99 rated pros aged 21 or younger at a potential of 90 or above. Europe has the most young pros in the archive, 295, thanks to its many national leagues, but only 3 of them reach 90.

A region is where the player's team competes, not their nationality: HongQ, from Taiwan, counts for China, and LazyFeel, from Vietnam, for Korea.

How the model was built

The potential model learns from the careers of every pro in the archive: for each player and month, what was known that month, and the best level that player reached in the two years that followed.

• What it predicts. The best three-month level in the next 24 months, for players who play at least three of those months.
• What it sees. Only what was known that month: the current level and its change over six months, raw form, age, games played, the share of them in a top league, the best level so far, months active, how uncertain the rating still is, how the player compares with pros of the same age, and the role. Nothing from later months.
• The model. Gradient-boosted trees (LightGBM), one for the middle of the range (potential) and one for its 90th percentile (ceiling). A separate set of the same trees gives the level one, two and three years ahead, the curve on each player's profile.
• The test. Trained on 42,183 player-months up to December 2021 and scored on 8,269 player-months from January 2024 (1,795 players). The two-year gap means no training answer overlaps a test month's future.

Age matters: without it the error grows by about 6%. An explicit aging curve was also tried as an input and added nothing the trees had not already learned from age.

How it did on past players

On the 1,795 players of the test, the model's potential missed the level they actually reached by 7.9 points out of 99 on average. The best simple rule, today's level plus the average rise, missed by 9.9. For players aged 21 or younger, the ones potential is for, the model missed by 9.7 against 11.4. 91% of players ended at or under their ceiling, close to the 90% it aims for.

The model also gives each player a chance of reaching the top tenth of their role within two years. It is not shown on the site: across all players it beats a simple baseline (the share of earlier players at the same level who got there), but for players aged 21 or younger it loses to that baseline on all five measures below, AUC by a hair.

The year-by-year curve, tested on months from January 2023, beats both carrying today's level forward and the typical curve for the player's role and age at every horizon. Its range aims to hold four players in five; it held between 70% and 76%.

What this can’t tell you

• Potential reads "if the player keeps playing". A player who retires or stops being picked has no future level, so the model never learned from them.
• The level is adjusted for league strength, and that adjustment is itself an estimate: an academy or small-league player's level is less certain than a major-league starter's.
• There is no minimum number of games: a player with few pro games (re0 has 28) has a less certain level, and the model takes that in through the rating's uncertainty and the games played.
• Few Chinese academy (LDL) players are rated in the archive, so China's count of young prospects is a floor, not a census.
• "None on record" is not "free to sign": only the leagues that publish contracts to Leaguepedia's database have dates, and a contract can be extended or bought out at any time.
• Solo-queue ranks change daily and say nothing about potential: they are shown for context, from the account Leaguepedia or LoLPros published.
• Every number is as of Nov 12, 2026. Levels and potentials are rebuilt every morning; the live list is on the Prospects and Scouting pages.