HLTV has updated Rating 3.0 to improve the accuracy of its player-performance calculations after collecting data from 29,030 Counter-Strike maps. The sample includes 9,029 maps played on LAN and allows the system to account more precisely for differences between individual maps.
The revision is not a replacement for Rating 3.0. It is a targeted recalibration of win probabilities, averages and Round Swing rules. Historical figures will move in some cases, but HLTV’s review of recent matches indicates that the typical change is small.
Map-specific context is central to Rating 3.0 because the same five-versus-five situation does not carry identical winning chances everywhere. HLTV noted that the probability can differ by almost ten percentage points between Anubis and a more CT-sided map such as Overpass, making a universal baseline less accurate for advanced statistics.
Cache exposed that limitation after entering the current game pool without enough map-specific information. The system initially used global averages for it. Current data shows that Terrorists win 47.1% of full-buy rounds on Cache, giving the calculation a dedicated baseline rather than treating the map as average.
What changed inside Rating 3.0
The update incorporates Counter-Strike results through September 29, 2026 into the win probabilities that support Round Swing and eco adjustment. HLTV also refreshed the averages and standard deviations behind every sub-rating with LAN data through the same date, preserving an event-wide average rating of one per player.
Several edge cases around planted bombs and late kills were tightened. A kill no longer contributes to Round Swing when fewer than ten seconds remain on the bomb, or fewer than five seconds when the Counter-Terrorists have a defuse kit, unless a CT is actively defusing. This prevents a statistically dramatic reward for a kill that cannot realistically change the round.
The published patch notes identify six practical changes:
- Win probabilities now use data through September 29 and include Cache-specific figures.
- Sub-rating averages and deviations were recalculated from current LAN data.
- Very late bomb-timer kills receive no Round Swing in defined situations.
- Excess post-plant swing credits only kills recorded after the bomb plant.
- A kill can no longer create negative Round Swing for the player.
- Stand-ins receive no Round Swing for rounds they did not play.
The restriction on negative Round Swing addresses an unusual mathematical outcome in which win probability could still fall after a kill. When that happens under the revised model, the player receives zero swing instead. The change favors a neutral result over penalizing an action that removed an opponent.
Post-plant distribution is also more selective. When the system assigns excess swing at the end of a planted-bomb round, only kills after the plant are considered. That ties the credit to events occurring during the state whose probability is being redistributed.
These refinements matter most at the boundaries of the model. They do not turn every scoreboard upside down, and Rating 3.0 still combines several components rather than functioning as a simple kills-and-deaths formula.
How much historical ratings will move
HLTV tested the update on 6,687 maps played from July through September 2026. The average movement in a player’s single-map rating was 0.021. In 9.4% of player-map records the change reached at least 0.05, while only 1% moved by 0.10 or more. The largest recorded adjustment was 0.31.
Cache is naturally the most affected map because it now has its own dataset. Inferno is the map most influenced by the revised bomb-timer handling due to its post-plant characteristics. Those effects follow the specific gaps the update was designed to address rather than representing a broad change in how every map is valued.
The recalibration arrives before the 2026 HLTV Awards and annual player ranking calculations. Greater precision is the stated purpose, but a rating remains a statistical summary, not a complete substitute for watching roles, tactics and match context.
| Measurement | Reported figure | Interpretation |
|---|---|---|
| Maps in database | 29,030 | Broader basis for probabilities |
| LAN maps | 9,029 | Current offline context |
| Test sample | 6,687 maps | July-September 2026 review |
| Average movement | 0.021 | Typical adjustment is limited |
| Largest movement | 0.31 | Rare upper-end change |
The Rating 3.0 update uses a much larger body of CS2 evidence to correct map baselines and a small group of unusual Round Swing outcomes. Cache gains its own context, late-round events are treated more carefully and stand-in participation is handled explicitly.
Most existing single-map ratings should change only slightly, according to HLTV’s test sample. The revision is therefore best understood as maintenance of the model before year-end awards, with the largest effects concentrated where earlier data or rules were weakest.
Comments