Radiant or Dire: which side wins more in Dota 2?
It’s one of the most common questions about Dota 2: is one side of the map better? We counted the results of 90,251 matches played between January 2025 and September 2026.
Radiant wins a little more often
Radiant won 52.3% of all matches. The gap is small, but it is consistent:
- in professional leagues - 52.6% (25,414 matches);
- in other leagues - 52.2% (64,837 matches);
- in every patch of the period: 51.4% on 7.37, 52.7% on 7.38, 51.7% on 7.39, 52.3% on 7.40 and 52.6% on 7.41.
With this many matches, two points is not a fluke. Over a thousand games, that is about 523 wins for Radiant and 477 for Dire.
The advantage is an early-game thing
When we split matches by length, the picture becomes much clearer:
- games under 25 minutes - Radiant wins 56.4%;
- 25–35 minutes - 53.0%;
- 35–45 minutes - 50.1%;
- 45 minutes or more - 49.8%.
In other words, Radiant’s edge shows up in games that are decided quickly. Once a match goes long, the sides are practically even.
Why?
The data tells us that Radiant wins more, not why. The explanations usually given by players and analysts are about the map: the layout of the lanes and jungle camps, which side the Roshan pit and the runes favour at different times, and even the camera angle, since Radiant attacks “up” the screen. Valve changes the map often, and the size of the gap moves a little from patch to patch, but in our period it never flipped.
Does it matter?
For a single game, hardly: 52% is very close to a coin flip, and team strength, the draft and the game itself matter far more. But across a tournament, two points add up, and in competitive Dota the side choice can be traded for draft priority.
Our models know about it too. Before anything else is known, the starting point is not 50/50 but about 52/48 for Radiant, and the live model also takes into account that the edge fades as the game goes on. You can see all three estimates for every live match on Swingmeter, and how long matches usually last in our article about game length.