HLTV has released a small accuracy update for Rating 3.0, using data from 29,030 CS2 maps played since the system was introduced. The changes affect map-specific win probabilities, Round Swing, eco-adjustment and some historical ratings, with 9,029 LAN maps included in the updated dataset.
The update was made ahead of the 2026 HLTV Awards and Top 20 Players of the Year, with the goal of making Rating 3.0 more accurate across different maps and game situations.
Rating 3.0 gets more accurate map-specific data
One of the main changes concerns the win probabilities used for Round Swing and eco-adjustment.
Rating 3.0 uses separate averages for each map instead of treating identical situations the same everywhere. In a 5v5 buy round on the T side, the win probability is 53.5% on Anubis, while on Overpass it drops to 44.2%.
Cache was one of the biggest gaps in the system because there was not enough CS2 data for the map when Rating 3.0 was released. HLTV therefore had to rely on global averages.
The map now has its own data, with the T side winning 47.1% of buy rounds. If there is still not enough information for a specific scenario, such as a particular post-plant or player advantage situation, Rating 3.0 will continue using global averages instead.
HLTV also updated the underlying averages with matches played after Valve changed the economy in July 2025, which made the game more CT-sided.
Round Swing receives several changes
HLTV also changed the way Round Swing handles rounds that have effectively already been decided.
Kills will no longer contribute to Round Swing when fewer than 10 seconds remain on the bomb and the CT side does not have a defuse kit. If the CTs have a kit, that limit becomes five seconds unless a player is actively defusing.
This means kills after the CT side can no longer realistically complete the defuse will be treated more like exit frags instead of producing a large Swing value.
Another issue has also been fixed. In some rare scenarios, a player could previously receive negative Round Swing after getting a kill because the calculated win probability of the team went down.
If that situation occurs now, the player will receive 0 Swing instead of a negative value.
HLTV also changed the distribution of excess Swing after bomb plants and removed Round Swing from stand-ins for rounds they did not actually play.
Historical ratings will change slightly
The updated data also required HLTV to recalculate averages and standard deviations used across Rating 3.0.
Across 6,687 CS2 maps played between July and September 2026, a single-map Rating 3.0 changed by only 0.021 on average.
According to HLTV, 9.4% of player-map ratings changed by at least 0.05, while 1% changed by at least 0.10. The largest individual change was 0.31.
Cache is naturally the map affected most by the addition of its own map-specific data. Inferno is also among the most impacted maps because of its high number of post-plant situations and the changes to how bomb timer scenarios are handled.
HLTV said the update should improve the accuracy of Rating 3.0 before the end-of-year awards and the 2026 Top 20 Players ranking.
