Hanabi (card game)
Cooperative card game where players see others' hands but not their own.
Hanabi is a cooperative card game designed by French creator Antoine Bauza and released in 2010. Players see everyone else’s cards but not their own, and they cooperate to lay down cards in a prescribed sequence, mimicking a fireworks display. The deck has five suits—white, yellow, green, blue, and red—with three copies of 1, two copies each of 2, 3, and 4, and one copy of 5. The game starts with eight information tokens and three fuse tokens. Each player receives a hand of five cards (or four in games with four or five players). On a turn, a player must either give information, discard, or play a card. Giving information uses one token and lets the player point out all cards of a specific number or suit in another player’s hand, and the information must be complete and correct. Discarding a card replenishes one information token, and the card is removed from play. Playing a card succeeds if it is a 1 in a suit not yet started, or the next number in a suit already played; otherwise, a fuse token is lost and the card is discarded. Successfully playing a 5 restores one information token. The game ends immediately if all fuse tokens are used (a loss) or if all 5s are played (a win). The game also ends when the deck runs out: after the last card is drawn, each player takes one final turn, and then scoring occurs. The final score is the sum of the highest card played in each suit, out of a possible 25 points.
Hanabi received positive reviews. Board Game Quest gave it four and a half stars, highlighting its uniqueness, accessibility, and engagement. The Opinionated Gamers also praised its engagement and addictiveness. It won several awards, including the 2013 Spiel des Jahres and the 2013 Fairplay À la carte Award, and placed sixth in the 2013 Deutscher Spiele Preis.
Computer programs can play Hanabi in self-play (multiple instances of the same program cooperating) or ad hoc team play (playing with other programs or humans). Hand-coded rule-based programs, like WTFWThat, have been reported to achieve high scores in self-play with five players, sometimes averaging around 24.9 out of 25 in specific controlled tests, though such results are not a universally recognized canon fact and depend on exact conditions. In 2019, DeepMind proposed Hanabi as a benchmark for AI research in cooperative play.
Quick Facts
- Title
- Hanabi
- Designer
- Antoine Bauza
- Publisher
- R&R Games, Cocktail Games, Abacus Spiele
- Players
- 2 to 5
- Ages
- 8 and up
- Setup Time
- 5 minutes
- Playing Time
- 20–30 minutes
- Random Chance
- Medium
- Skills
- Deduction, memory, cooperation, planning
Facts from the source article.
Lore & Background
Hanabi was published in 2010 and quickly gained attention for its unique cooperative gameplay. Players are dealt a hand of cards they cannot see, while being able to see the hands of all other players. The goal is to play cards in ascending order across five suits to create a fireworks display. Communication is limited: players may only give information about the number or suit of cards in another player's hand, and each such action consumes one of eight information tokens. Discarding a card replenishes a token, and successfully playing a 5 also restores one token. The game ends when all three fuse tokens are lost (a loss) or all five 5s are played (a win), with scoring based on the highest card played in each suit, out of 25 points.
The game received positive reviews, with Board Game Quest awarding four and a half stars and The Opinionated Gamers praising its engagement and addictiveness. It won the 2013 Spiel des Jahres and the 2013 Fairplay À la carte Award, and placed sixth in the 2013 Deutscher Spiele Preis. Hanabi has also become a benchmark for artificial intelligence research, particularly in cooperative play and theory of mind.
Reader's Guide
Hanabi's significance lies in its innovative cooperative mechanics and its role as a benchmark for AI research. The game requires players to reason about the beliefs and intentions of others, as they can only give limited, complete information about other players' hands. This makes it a rich domain for studying ad hoc team play, where AI agents must adapt to unfamiliar partners. In 2019, DeepMind proposed Hanabi as an ideal game for establishing a new benchmark for AI research in cooperative play. Self-play programs have achieved near-perfect scores (24.9 out of 25) using hand-coded strategies, while learning-based programs initially scored only about 15 points but improved to around 24 by 2020 with the Simplified Action Decoder. Ad hoc team play remains a greater challenge, as programs must learn communication conventions and strategies with other agents or humans. Hu et al. showed that learning symmetry-invariant strategies improved performance with separately trained AI agents (scoring around 22) and with humans (scoring around 16). DeepMind released the Hanabi Learning Environment, an open-source framework to facilitate research. The game thus bridges recreational gaming and cutting-edge AI development.
Did You Know?
- Hanabi won the 2013 Spiel des Jahres, an industry award for best board game of the year.
- The game ends immediately when all fuse tokens are used up, resulting in a game loss, or when all 5s have been played successfully, leading to a game win.
- The best hand-coded computer programs achieved near-perfect results in self-play with five players, averaging 24.9 out of 25 points.
More in Cooperative board games 1-24
Spotted an error? Know more?
Reader corrections go straight into our review queue. Suggest an edit · How this site is sourced