In ranked Valorant, Haven is defender-sided. Across 65,323 matches on MetaBot’s Haven tracker (all ranks, October 7, 2026), the defending side wins 52.1% and the attacking side 48.8%. That 3.3-point gap is the same one we recorded (52.2% vs 48.9%) in our Haven map guide in July, so it has held for three months.
The “attacker-sided” label isn’t invented, though. It comes from pro play, where Haven was among the most attack-friendly maps at Champions 2025. And the usual explanation, that a third site stretches defenders too thin, fails a simple test: Lotus is the only other three-site map in the pool, and it leans the other way. This article separates three questions: who wins on Haven, whether the third site is the reason, and where defenders really do get stretched.
Verified on Patch 13.06 (September 22, 2026). Neither the 13.04 nor the 13.06 patch notes list a Haven change, and Haven stayed in Competitive when 13.04 swapped Abyss in for Breeze. Win rates move every patch, so treat every figure below as a dated snapshot.
Quick Start: The Haven Side Question in Six Lines
- Ranked verdict: defense wins 52.1% to attack’s 48.8% (MetaBot, 65,323 matches, October 7, 2026).
- Pro verdict: attack-sided at Champions 2025 (57.73% attack win rate), which is where the reputation comes from.
- The third site isn’t the cause: Lotus has three sites and favors attackers; Ascent, Split and Summit have two and favor defenders harder than Haven does.
- On defense: anchor B with one player and spend your second body and your utility on A and C, the two sites with two entry routes.
- On attack: take Mid first, then fake one site and hit another. A five-stack at a single door is the version of Haven that loses.
- Queue check: Haven is still in the Competitive pool; the rotation details are in our 2026 map pool and Act rotation guide.
Ranked Haven Is Defender-Sided; Pro Haven Was Not
Ranked defenders win more on Haven, and the lead has been stable across a growing sample. Pro play in 2025 looked different. The table puts the three datasets side by side.
The takeaway in one sentence: ranked data says defense by about three points, one pro event said attack by about fifteen (57.73% against an implied 42.3%), and neither is wrong because they measure different players.
| Dataset | Sample | Attack win rate | Defense win rate | Reads as |
|---|---|---|---|---|
| Ranked, all ranks, Oct 7, 2026 (MetaBot) | 65,323 matches | 48.8% | 52.1% | Defender-sided |
| Ranked, all ranks, July 2026 (our Haven guide, MetaBot) | 45,138 matches | 48.9% | 52.2% | Same split |
| Pro, Champions 2025 (The Spike) | One event | 57.73% | Not stated (implies about 42.3%) | Attacker-sided |
Two cautions on that table. MetaBot’s two Haven figures add up to 100.9%, so they are not a clean round-by-round split; read the gap, not the individual numbers. And the Champions figure comes from The Spike’s Champions 2025 map analysis, a single tournament played before Haven’s Patch 12.00 Mid Window panel change, so it is a hint about pro play and not a verdict.
Why the “attacker-sided” reputation survives
The reputation was written in a pro-play era and by people who watched coordinated teams. A 2023 PCGamesN piece on Haven and Lotus called Haven “a heavy attacker-sided map” and said the layout, not the three sites, was the problem. Its example was the fake-A, go-C default: “One go fake A – rest go C… it works like a charm every single time.”
That default still works for teams that run it well. At Masters Santiago, G2 Esports picked Haven four times and won all four, posting a 76% attack-side win rate with a double-Duelist comp (Yoru and Phoenix) against only 48% on defense. That is one team’s record, not the map’s. But it shows what an organized attack does to a stretched defense.
Our reading, which is an inference and not a measured result: ranked attackers mostly don’t run that fake-and-split cleanly, so the stretch exists but rarely gets cashed in. We could not find a public dataset that splits Haven’s win rate by rank bracket, so we can’t prove it. If you want to test it yourself, MetaBot lets you filter Haven from Iron to Radiant.
The Third-Site Test: Haven vs Lotus vs the Two-Site Maps
A third site doesn’t decide which side wins, because the seven maps in the current pool split on both sides of Haven regardless of site count. Here is MetaBot’s defense-win ranking (September 26, 2026, all ranks), filtered to the seven maps in Competitive since Patch 13.04.
| Map | Bomb sites | Defense win rate | Attack win rate | Defense edge (points) |
|---|---|---|---|---|
| Ascent | 2 | 54.7% | 45.8% | +8.9 |
| Split | 2 | 53.0% | 47.8% | +5.2 |
| Summit | 2 | 52.8% | 47.5% | +5.3 |
| Haven | 3 | 52.1% | 48.7% | +3.4 |
| Sunset | 2 | 51.9% | 48.5% | +3.4 |
| Lotus | 3 | 49.7% | 50.8% | -1.1 |
| Abyss | 2 | 49.6% | 51.4% | -1.8 |
The simple average defense edge of the five two-site maps is +4.2 points; Haven’s is +3.4. So the third site costs defenders, at most, under one point on average. Meanwhile the two-site maps themselves run from +8.9 (Ascent) to -1.8 (Abyss), a spread of 10.7 points. Map geometry matters enormously; the site count barely registers.
Lotus is the cleanest control. It has the same three sites, per MetaBot’s Lotus page (50.7% attack, 49.7% defense across 39,381 matches on October 7), and it favors attackers. Our Lotus map guide covers its layout and site routes in detail. (Our older Haven guide calls Haven the only three-site map in the pool; that was wrong, and Lotus is the correction.)
The pro data points the same way. At Champions 2025, two-site Abyss posted a 57.93% attack win rate, essentially level with Haven’s 57.73%, while three-site Lotus sat near balance at 51.04%. Two datasets, two skill levels, and the same lesson: three sites alone don’t make a map attacker-sided or defender-sided.
One honest limit: with only two three-site maps in the pool, this shows the third site isn’t the dominant factor. It can’t prove the third site does nothing.
Where Defenders Really Do Get Stretched
Defenders get stretched in two specific ways: fewer bodies per site, and a long rotation between A and C. Neither one decides the side on its own, but both explain why Haven punishes a defense that guesses wrong.
The body count is plain arithmetic. Five defenders across three sites is about 1.7 players per site; across two sites it is 2.5. The map’s own design pitch frames the extra ground differently, though. Riot’s official Haven description says “there’s more territory to control, but defenders can use the extra real estate for aggressive pushes.” In other words, the third site is meant to give defenders room to take space, not only more to guard.
The rotation is where the stretch becomes real. Rotating between A and C is the long way around the map, and the PCGamesN piece called it “a nightmare unless you’re running on Brimmy’s Stimmy.” Dodge.gg’s 2026 Haven guide says the map “slightly favors defenders when they maintain Garage control and rotate efficiently,” which is the same idea from the other direction: lose Garage or Window and the rotation fails.
We did not find a measured A-to-C rotation time in any published source, so we won’t print one here. Our Haven guide’s “roughly 10 seconds once Window is lost” is a practitioner estimate and should be read that way. What you can rely on is the structure: on a two-site map, one short path links the sites; on Haven, the far site is a full-map trip.
Site by Site: What Attackers Get and What Swings the Round
Each Haven site asks a different question of the defense, and the answer depends on how many ways in it has. No public dataset we found breaks win rate down by site, so this table uses entry routes and setups from our Haven map guide rather than invented site percentages. Dodge.gg calls B the most commonly attacked site, but we found no data behind that claim.
| Site | Routes in | What attackers get | Defender setup that swings it |
|---|---|---|---|
| A | 2 (A Long, A Link/Lobby) | A real split: one lane pins the anchor, the other takes the site | Two players, or one plus an alarm or trap on the flank (Cypher, Killjoy) |
| B | 1 (Mid Courtyard into B Short/Window) | Little, unless Mid is already won | One anchor and good ears; spend nothing extra here |
| C | 2 (C Long, Garage/Tunnel) | The classic fake-A, go-C target | Two players, or one plus a chokepoint tool at Garage (Viper wall, Deadlock barrier) |
Read that table as a budget. B is the one site a lone defender can hold, so the extra body and the extra credits belong on A and C. Teams that split their five evenly, 1-1-1 with two roamers, pay full price for B and run short on the sites that actually have a second door. For the buy math behind that, see our Valorant economy guide.
Reading the Stretch in Real Time: A Defender’s Decision Table
A stretched defense loses when it rotates on the first noise, because on Haven a wrong rotation means a full-map trip back. The table below is a set of general guidelines drawn from the site structure above, not tracked statistics; adjust it to your rank and your team’s habits.
| What you see or hear | What it usually means | What to do |
|---|---|---|
| Utility on A Long or C Long, no bodies | A fake, or a lane being cleared before a split | Hold your site; don’t pull the B anchor or the far-site player yet |
| Window or Garage is lost | Your rotation path is now contested | Commit each player to the site they can hold and plan the retake, not the rotate |
| Info shows four attackers on one side | The real site; the fifth is the fake | Rotate only the second player from the nearest site, and call the fifth’s location |
| Quiet on all three sites for a long stretch | A slow default, often a split in progress | Stay put; use a trap, camera or recon dart to buy the read |
The logic is simple: every rotation spends time you don’t get back, and a fake exists to make you spend it. A defense that rotates only on confirmed information makes the fake cost the attackers a player’s worth of tempo and costs the defense nothing. That discipline, more than the number of sites, is how a stretched defense gets its time back.
What the Agent Data Says About Holding Three Sites
The agents winning most on Haven are the ones that watch ground without standing on it. Per MetaBot’s Haven agent table (October 7, 2026), Killjoy leads at 53.4% win rate (5.0% pick rate), followed by Phoenix at 53.3%, Neon at 53.1% and Clove at 52.9% (15.3% pick rate). Cypher is fifth at 52.0%.
Two Sentinels in the top five fit the stretch problem: a turret or trap covers a flank when you can’t. See our Killjoy guide and best Sentinel comparison. Phoenix and Neon suggest the other half, fast re-positioning, though that link is our inference and not something the tracker states. Chamber, the most-picked Sentinel at 7.5%, sits at 48.9%, so a popular pick isn’t automatically a strong one here.
On attack, information is what beats the stretch. A recon bolt that confirms where the extra defender went is worth more than another flash, which is why our Sova Haven lineups cover all three sites. And since MetaBot’s numbers are a community tracker with undisclosed sampling, the gaps between agents of a point or two are worth checking in your own games before you rebuild a comp around them.
What to Do by Player Type
The right Haven plan depends on whether you queue solo or with five, because coordination is the variable the data can’t see.
| Player type | On defense | On attack |
|---|---|---|
| New player | Hold B alone and call what you see; don’t rotate on sound alone | Stay with your team and trade; don’t peek C Long or A Long solo |
| Casual solo queue | Pick a Sentinel or Clove so a flank is covered without a teammate | Ask for one player to fake a second site before the team commits |
| Hardcore / optimiser | Track Window and Garage control each round and rotate on utility use, not footsteps | Default to the fake-A, go-C split, then vary the fake so the defense can’t pre-rotate |
| IGL / five-stack | Run 1 on B, 2 on A, 2 on C, with a reserved late-round rotation caller | Treat a split like G2’s: commit two to a flank lane, three to the real site, and use clear callouts |
If you only take one rule from the decision tree: on defense, ask “did we lose Window or Garage?” If yes, stop rotating late and commit to the site you can hold. On attack, ask “has the defense shown a rotation?” If no, fake and go; if yes, hit the site they just left. Our post-plant guide covers what to do once the spike is down, and the new player guide is the place to start if Haven is still a maze.
FAQ
Is Haven attack or defense sided in Valorant?
Defense-sided in ranked, by about 3.3 points (52.1% to 48.8%) over 65,323 matches. The pro scene looked attack-sided at Champions 2025, so the answer depends on who is playing.
Why do people say Haven is attacker-sided?
Because coordinated teams can punish a split defense with a fake-A, go-C execute, and pro results like G2’s 76% attack record at Masters Santiago make that visible. Ranked attackers rarely execute it that cleanly, which is why the label doesn’t carry over.
Does having three sites make Haven harder to defend?
Slightly at most. Haven’s defense edge (+3.4) is under one point below the two-site average (+4.2), and Lotus, also three-site, favors attackers. Map layout moves win rates far more than site count does.
Which Haven site is easiest to defend?
B, because it has one entry route and a single anchor can hold it. A and C each have two, so they need a second player or a utility tool covering the flank.
Is Haven still in the ranked map pool?
Yes. Patch 13.04 swapped Abyss in and Breeze out, and neither 13.04 nor 13.06 touched Haven. Check the in-game pool before each Act, since rotations land on Act boundaries.
Sources
- MetaBot.GG. Valorant Haven map overview, all ranks (October 7, 2026); Valorant Lotus map overview (October 7, 2026); Valorant defense-win map ranking (September 26, 2026); Valorant Haven agent statistics (October 7, 2026).
- Riot Games. Haven. VALORANT Wiki (bomb sites, official description, destructible panels).
- Riot Games. VALORANT Patch Notes 13.04 (August 18, 2026; Abyss in, Breeze out).
- Riot Games. VALORANT Patch Notes 13.06 (September 22, 2026; no Haven change).
- The Spike. VALORANT Champions 2025 map analysis: trends and insights (October 8, 2025); The best team on every map at VALORANT Masters Santiago (March 18, 2026).
- PCGamesN. Valorant’s triple-site maps: Haven and Lotus, Fariha Bhatti (April 10, 2023).
- Dodge.gg. Best Agents & Weapons for Haven (2026), Valorant Map Guide, March 21, 2026.
Win-rate figures are community-tracker and press data, not Riot-published statistics, and change with every patch. Verified against Patch 13.06 on October 7, 2026. The diagrams in this article are schematic illustrations, not Haven’s actual layout.
I've been playing video games for over 20 years, spanning everything from early PC titles to modern open-world games. I started Switchblade Gaming to publish the kind of accurate, well-researched guides I always wanted to find — built on primary sources, tested in-game, and kept up to date after patches. I currently focus on Minecraft and Pokémon GO.