Crazy Stone is a strong program by Rémi Coulom which uses Monte Carlo Tree Search and pattern learning. Crazy Stone came second in the 2007 Computer Olympiad behind MoGo. MoGo claims they were partly inspired by Crazy Stone (they are both French).
In spring of 2011, Crazy Stone, playing as 'bonobot', briefly reached 5 dan on KGS, and maintained a 4 dan rating.
On December 2nd, 2007, Crazy Stone won both of the KGS tournaments and the UEC Cup in Japan, ahead of MoGo in all three events. In 2008, Crazy Stone defended its UEC Cup title, defeating Many Faces of Go. Afterwards, it defeated Kaori Aoba, Japanese 4p, in a 7-stone handicap exhibition game. That gave Crazy Stone a claim to being the strongest program in the world at that time.
In March 2014, Crazy Stone established a 6-dan rating on KGS.
- Crazy Stone home page.
- Article in Wired magazine dated 19 September 2006.
- Second Article in Wired magazine dated 13 May 2014.
Crazy Stone is mentioned in the French public radio show Science Publique at France Culture on 2008-01-11 (there is a MP3 of the show).
An interview with Remi Coulom (June 2011): http://rusgo.org/crazystone/
In 2016, a new version that uses deep learning was able to significantly increase Crazy Stone's strength.
Since April 2011 : Crazy Stone is available as commercial product in Japan. It is sold as 最強の囲碁 by Unbalance Corporation.
An English version, Crazy Stone Deep Learning, is available at http://www.unbalance.co.jp/igo/eng/
The smartphone/iPad versions offer in-app purchases of tickets to play against the CrazyStone Engine Server which is significantly stronger than native app version.
The engine server version offers 10 levels of play. Currently, there are no details publicly available regarding the strength of the server version or the hardware on which it runs. However, according to some user feedback, it seems to be significantly weaker than KGS 6d as of 24 April 2016.
The in-app description explains:
"Crazy Stone is a popularly known Go engine, based on the Monte-Carlo evaluation technique, combined with tree search, and has achieved 6d at KGS rating. We have optimized this version for Engine Server games."
Some users claim that the engine server version has been "optimized" to play faster but weaker, and estimate it to play at around 1-3d KGS level.