7 Wonders Duel: which wonder is best, and is going first unfair?

Why draft data can't settle it

Every 7 Wonders Duel game starts with a draft: four wonders on the table, the first player takes one, the second player two, the first player gets the last. Then four more, with the second player picking first. Ask players which wonder is best and you get tier lists, often with the Temple of Artemis (12 coins and an extra turn) at the top and the Statue of Zeus or the Pyramids at the bottom.

There is real data too. In 2022 a Board Game Arena player collected about 5,000 games from the top 500 players: Piraeus won 54.1% of the games it was in and the Sphinx 53.1%, and the first player won 55%. That is better than opinion, but it has two problems. A wonder's win rate depends on who picks it: if the first player tends to take it, it inherits the first player's edge. And it depends on what it was picked over: taking the best wonder of a round means you also end up with that round's leftover.

A simulation can remove both. We dealt the wonders at random instead, four to each player, so every wonder lands with the first and the second player equally often and next to every other wonder. Then the win rate of the player who holds it is a clean measure of what the wonder is worth.

Piraeus comes out on top, as it did on Board Game Arena: 58.9% for the player who holds it. It costs only wood, clay and stone, the resources you have in Age I, gives an extra turn, and then makes a glass or papyrus for you every turn for the rest of the game. Our simple bot, which plays without looking ahead, ranks it first too (58.1%), and the two bots disagree by more than three points on one of the twelve wonders (Great Library).

The Temple of Artemis is third (53.9%): good, not the best. Twelve coins and an extra turn is a strong start, but coins are only worth something until you have enough of them, and the Temple scores no points of its own.

Why is the Great Lighthouse second?

Second place goes to a wonder nobody ranks highly: the Great Lighthouse (57.8%). It has no extra turn and no clever effect. It is 4 points and one wood, clay or stone every turn. That sounds dull until you count what it saves. Most Age II and Age III cards need raw resources, often two or three, and every one you lack costs 2 coins plus one for each your opponent produces. A player without stone facing a Shelf Quarry pays 4 coins per stone, every time. The Lighthouse quietly removes that problem for the rest of the game.

We trust this one a little less than the rest, for a reason we explain under How we tested: our bots win mostly on points, and in a points game a resource every turn is worth more than in a game decided by science or the military track. ZeusAI, a research AI that taught itself 7 Wonders Duel, dislikes the Lighthouse in its drafts.

Extra turns: how much are they worth?

Five wonders give an extra turn: Piraeus, the Temple of Artemis, the Hanging Gardens, the Appian Way and the Sphinx. As a group they win 52.2% against 48.5% for the other seven. The cleaner way to see it is to count them. A player dealt no extra-turn wonder at all wins 41%; one with three of them wins 56%. That holds for both bots.

Why an extra turn is worth so much: in a pyramid of cards you normally get every other card, and your opponent sees which ones you open for them. A second move lets you take two cards in a row, typically one you need and the one it was covering, and it flips who gets the next good card for the rest of the round.

One extra-turn wonder breaks the pattern: the Sphinx (47.9%, rank 8 of 12). On Board Game Arena it is the second best wonder. It needs two glass, and only two cards in the game make glass. More telling: the wonders that are mostly points do badly in our games. The Pyramids (9 points and nothing else) finish near the bottom, while the top three are all about economy: a resource every turn (Piraeus, the Lighthouse) or a pile of coins early (the Temple). In our games, an early economy beats points you could have had.

Which wonder is the worst? Not Zeus.

The bottom of the table is crowded. The Mausoleum (43.9%), the Pyramids and the Colossus are within a point of each other, which is less than our margin of error, so call it a three-way tie for last. Each has a clear weakness. The Mausoleum is one of the most expensive wonders and only as good as the discard pile, and it gets built in just 66% of games. The Pyramids are 9 points and nothing else. The Colossus is 3 points and two shields, which rarely decides a game on its own.

One claim you will find online does not survive: that the Statue of Zeus is by far the worst wonder, worse than Circus Maximus because Circus Maximus destroys the rarer resource. In our games it is the other way round. Zeus wins 50.7% and Circus Maximus 46.3%. Destroying a brown card hurts more than it looks: it costs your opponent a resource they built their plan on, and if you buy that resource, your price drops by a coin at the same time.

Why do draft stats look so even?

So far no one chose anything. In a real game the wonders are drafted, so we also played 3,000 games with the official draft, both bots picking by the values above, and measured win rates the way Board Game Arena data does: per wonder, over the games it was drafted in.

The draft squeezes the differences. Between the best and the worst wonder the gap shrinks from 15 points when dealt to 6 points when drafted. The reason is the leftover: in each round of four, whoever takes the best wonder also ends up with the one nobody wanted, and whoever is stuck with the worst one got the second and third pick with it. Draft win rates measure a wonder plus its company. That is why the Board Game Arena numbers sit so close to 50%, and why they understate how much the top wonders are worth.

How big is the first-player advantage?

The first player picks first in the draft and starts Age I. Between our strongest bots, with the official draft, the first player wins 53.7% of 7,200 games (± 1.2%). Our simple bot gets 55.7%, and top players on Board Game Arena 55%. So going first is worth about 4 points of win rate: real, but about half of what Piraeus is worth over an average wonder.

It does not come from the first pick. When we deal the wonders instead of drafting them, the first player still wins 55.0%. The edge is the tempo of starting Age I. Bots that play at random barely have it (53.2%); every bot of ours that plays with a plan lands between 54% and 56%, including a searcher that thinks three times as long (on fewer games, ± 4.6%). The one big outlier is ZeusAI, a research AI that beat top Board Game Arena players in 26 of 38 games: against itself, its first player won 66.8%. Perhaps the edge grows at the very top. At the level most of us play, it is four or five points.

Should you start Age II yourself?

Before Ages II and III, the player who is behind on the military track decides who starts (with the pawn in the middle, the player who took the last card decides). A common worry is that starting Age II is a bad deal: in its upside-down pyramid, the first card you take always turns over a card for your opponent. We forced the choice both ways in 1,500 games. When the player who chooses starts the age themselves, they win 56% of games; when they hand the start to their opponent, 38%. For Age II alone it is 61% against 44%. The card you open is a smaller gift than an age of moving first. Left to decide for itself, our strong bot starts the age 93% of the time.

Does a house rule fix it?

A common fix is to give the first player fewer coins. One house rule that has been around for years: the first player starts with 5 coins instead of 7, too few to buy two 3-coin yellow cards in Age I. We tested 7, 6, 5, 4 and 3 coins for the first player, against both bots, plus the variant ZeusAI's authors tried (the second player starts with 10).

Against our simple bot the rule does what it promises. Every coin the first player gives up costs them about 1.5% of win rate: with 5 coins they win 53.3% instead of 56.1%, and the game is even at 3 coins. Giving the second player 10 coins instead lands at 52.0%.

Against our strong bot, the rule does nothing we can measure. With 7 coins its first player wins 53.7% of 7,200 games, with 5 coins 55.4% of 4,200: if anything, slightly more. The difference is inside the margin of error. The other coin counts, with fewer games each, show at most a gentle slope: even starting with 3 coins, the first player wins 52.2%. A player who plans well loses little from two missing coins, because the first player's edge is not money: it is tempo. ZeusAI's authors found far more for their AI, about 2.7 points per coin, and still could not make the game fair with coins alone.

So does the house rule make the game fairer? Against our simple bot, yes: 5 coins takes away about half of the first player's edge, 3 coins all of it. Between our strong bots we could not measure any effect, because coins don't buy back the tempo of starting Age I. Try it at your table if you like; just don't expect it to settle the matter. The free fix costs nothing at all: when you get to choose who starts Age II or Age III, start it yourself.

Get the game: 7 Wonders DuelTwo players, about 30 minutes. Affiliate link.

Check the price

Already own it? 7 Wonders Duel tops our best 2-player board games ranking. The rest of that list is what to play next.

How we tested

The rules engine plays the base game of 7 Wonders Duel: the three card pyramids with their face-down rows, all 66 Age cards and 7 guilds (3 random cards removed per Age, 3 guilds shuffled into Age III), the 12 wonders with the official draft and the rule that only seven wonders can be built, the 10 progress tokens, trading at 2 coins plus the opponent's brown and grey production, the reserve and Customs House discounts, chains (and Urbanism), the conflict pawn with its looting tokens and military supremacy, pairs of science symbols and scientific supremacy, the weaker player choosing who starts Ages II and III, and the tie-break on blue points. 127 automated checks cover these rules, from which cards a pyramid opens to the Mausoleum taking a card the Statue of Zeus destroyed.

Three kinds of bots played. A random bot. A simple bot that scores every move one step ahead with a value for points, coins, resources, shields and science symbols, with weights tuned in 1,008,000 games against itself. And a strong bot that searches: for each move it redraws the cards it cannot see (the face-down cards and the later Ages), plays the game out to the end many times, and keeps the move that wins most often. It beats the simple bot in 90% of games. Wonder values come from the games with dealt wonders; the bots then draft by those values in the games with a draft. Every game runs from a fixed seed, so all numbers reproduce exactly.

Limits worth knowing. Our bots win mostly on points: 87% of games between strong bots end in a civilian victory, 12% on the military track and only 2% by science, where good human players on Board Game Arena reach about 58%, 16% and 26% (figures from the ZeusAI paper). Our bots under-use science, so wonders that shine in science-heavy games (the Great Library with the Law token, the Mausoleum bringing back a green card) may be worth more between humans than they are here, and resource wonders like the Great Lighthouse less. The strong bot is good, not perfect, so this is the best of the strategies we tested, not optimal play. We only tested the base game: no Pantheon or Agora. We typed the card costs in ourselves; if you spot one that is off, tell us and we will rerun it. More on how we test.