Inside PokerSnowie's brain reveals the work of the Snowie AI Team. It explores first hand how the brain of PokerSnowie evolves and learns advanced strategic concepts, on its own.

PokerSnowie's ultimate aim is to produce the perfectly balanced game, find the ultimate un-exploitable equilibrium for all No Limit Hold'em configurations. Join us on this fascinating journey, which is just starting, into the future of poker.

The PokerSnowie Video Quiz series aims to answer the question: "what is the right play?" It is a set of poker coaching videos presented by French professional poker player Sharp. In each video, Sharp sets up an interesting hand in the "Scenarios" tool of PokerSnowie and explains how to analyse the situation and learn from PokerSnowie's advice, based on the Game Theory Optimal model.

Challenge PokerSnowie - approaching 1 million hands played!

Challenge PokerSnowie update

There has been no Christmas break for PokerSnowie (although we will admit that the human beings looking after it were probably not quite as diligent as usual in the past few days..)!   

In any case, we're fast approaching 1 million hands played against PokerSnowie, and so it's time to provide you with an update.

Overall figures stand at 979,195 hands played, an average error rate of 17.31 and a global winrate for PokerSnowie of  22.85 bb/100 


To be consistent with our previous reporting,  we removed anything under 100 hands played, and any play with an error rate above 25.


  • 826,727 hands played
  • An average error rate of 14.55 which we classify as advanced.
  • PokerSnowie made a profit of 15.51 bb/100

How did the results by Error Rate evole? Interestingly, with more hands played, we now see a clear correlation between win rate and error rate. In fact PokerSnowie is now losing against its better opponents. (Note that the Extra-Terrestrial level does not have anywhere near enough data to be statistically significant)

PokerSnowie's win rate by Error Rate level
  Hands played Error rate PokerSnowie's win rate
(in bb/100)
Extra Terrestrial
1,217 3.10 -37.01
World Class
75,684 6.61 -2.78
187,069 9.78 +5.02
222,605 13.65 +13.34


135,175 21.79 +37.96


Since Challenge PokerSnowie offers the possibility to play from Head's Up to Full Ring, we thought it would be relevant to show the results by types of games. (We only report below on the games played at $0.5/$1, which represent 70% of the hands played against PokerSnowie and seem to be representative of the overall results.)

Challenge PokerSnowie results by game type
  Hands Played PokerSnowie's win rate
(in bb/100)
Head's Up 227,461 +19.6
6-Max 228,246 +13.5
Full Ring 65,622 +19.9


And what about Serious challengers who played in excess of 5000 hands against PokerSnowie?

23 players raised to the challenge so far, playing a total of 184,198 hands. Their average error rate is an advanced 15.44. And they lose in aggregate 14.74 bb/100.

However, amongst those, 7 players did manage to win against PokerSnowie, with win rates in bb/100 going from 0.2 to 18.27 for the biggest winner. 2 of the winning players went head's up against PokerSnowie whilst the other 5 found their edge at a 6-max table.

If you want to be acknowledged by name as a Challenge PokerSnowie winner (5000 hands minimum), please let us know by email or in the comments.


Till the next update, we wish you all a very happy end to 2013. Thanks a million times for your support and interest this year, and we look forward to continuing the quest to perfect GTO poker with you in 2014.


Snowie is not indestructible, it's a good news, the human brain have yet good days!
Hey, i play seriously my last 5018 hands in challenge snowie to get a positive winrate. And i'm happy to see my results. My winrate during these hands is to +36bb/100. My error rate is 9.84. I think i can play better. I just try some stupid moves to see how snowie react. Thank you for your job and your soft.
Hi Ludovic, well done indeed; about 6300 hands, with an overall win rate of 17.71 bb/100, error rate of 17.9. Glad you like PokerSnowie, we hope you benefit from it.
In your next blog post, please feel free to acknowledge my victory for mankind against our future robot overlord! PokerSnowie is definitely the best bot I've played against and gave me a run for my money. Getting chips from that thing is like trying to take meat from an alligator.
Hi Jay, acknowledgement coming right away - you played at an impressively low error rate of 7.11, for an overall win rate of 2.98 bb/100. Do you think you can keep your result for an other 10 000 hands? :-) Well done!
A user called fontaine has posted on the 2+2 Pokernowie game theory thread that they beat Pokersnowie for about 35bb/100 over about 6k hands. They are playing very lose and aggressive to take advantage for Pokersnowie folding to much. But you won't put them on the list because they have an error rate above 25, and was not playing 100nl.
Hi Andy, Yes, we do indeed have a new "champion" player, who managed to win 36.29 bb/100 over 6000 hands, and an error rate of 25.48. 35 players in total have played in excess of 5000 hands. Overall results are: - 282,387 hands played by 35 players; - PokerSnowie won 10.9 bb/100; - Global error rate is a pretty strong 15.09; - 12 players have managed to win against PokerSnowie, including 2 players who played in excess of 10,000 hands each; - Biggest win rate is as mentioned above, 36.29 bb/100; second biggest 18.27 bb/100; - The two best error rate amongst winners are 7.11 (for 5,004 hands) and 7.36 (for 10,384 hands).
:) I am the player who won 36 bb/100. I am curious to know what are the thoughts of the Snowie team about my results playing such an outlandish style? I certainly felt that I was able to gain an edge on Snowie. Do you agree, or do you think I had top 1% rungood?
Hi Kjetil, We'll have to agree that this unusual style of play is currently exploiting a leak in PokerSnowie's strategy. However, it's also fair to say that this weakness doesn't have an effect on PokerSnowie's analysis of regular games, as there these situations would hardly ever happen. We are looking at how to plug this leak though - and may have to invite you for another challenge soon :-)
Hi Kjetil, Are you up for another challenge? We think we've corrected the behaviour of PokerSnowie and that your min bet exploitative strategy now would not work anymore.
Do you think that fontaine's success shows that PokerSnowie is not yet close to GTO? For reference, a short description of the highly exploitative strategy he used is here:
My question is slightly different: If a player has learned to play a GTO game (perhaps by using Snowie), should that player deviate from GTO play when then see an opponent doing something that is clearly NOT correct? In short, would the developers of Snowie selectively use exploitative play against an opponent with a "leak?" I recognize, of course, that this type of play (deviating from GTO) opens one up to being exploited themselves. So, the question becomes: Always play GTO? Or deviate from GTO into exploitative play when it seems to be appropriate? Thanks...Terry
Hi Terry, If you play against a weak opponent and you know his exact weaknesses, you can certainly win more against him by deviating from GTO. The best is if you know the GTO strategy in the first place, so that you know the base from where to deviate.
(sound of PokerSnowie programmers quickly working up a sparbot following the exploitative strategy from the 2+2 forum). If each successful challenger helps develop a sparbot that beats PokerSnowie until it learns how to deal with that counter-strategy, would PokerSnowie eventually evolve into an unbeatable GTO? Or will PokerSnowie have to profile its opponents and switch to an "Anti-Fontainebot" or "Anti-Jaybot" if necessary. It will be very interesting to watch its evolution and see what path it takes toward world-domination like Snowie did for Backgammon.
Hi again Jay, during PokerSnowie's training all kinds of counter-strategies are automatically generated and PokerSnowie learns to defend against them. So the user's input is normally not required. "Fontaine", however, used a strategy which by design could not be generated during training. In this special case the input from Fontaine is very helpful, as now we can fix that problem. Thanks for your comments.
I'm curious as to why you were (/are?) so confident in Snowie's training if, by design, that training couldn't generate the sort of highly exploitative strategy 'Fontaine' employed. Suppose you plug this leak. Do you have reasons for thinking there won't be more (perhaps many more) exploitative strategies that will beat Snowie? (I don't mean to carp, but you are selling a product on the back of some pretty strong claims - Snowie is probably best NL player in world, Snowie is playing close to GTO etc - so there are ethical considerations here too)
Hi John, one important thing to note here is that Fontaine's strategy was highly unrealistic for any normal game - so clearly that type of exploitative strategy cannot be a priority for PokerSnowie. That said, we will fix this going forward. To try and and answer your overall question, by design, PokerSnowie only has 3 different bet sizes. The training has catered for bet sizes in between those bet sizes, so PokerSnowie would still play well against a bet size of 0.7 pot, for example. Potential problems are bet sizes that are either very small (like in the Fontaine case) or very big (like 10 times pot or more). With very big pot sizes no problems are known until now though. For more on some of PokerSnowie's weaknesses, please read this: Hope that answers your question.
Hello Team! Im new to this web page and I wanted to say Hi. This can be a great web-site and im glad I joined. New to this blog thanks for the welcome. I just came to this wonderful blog and wanted to introduce myself to everyone. This really is such an excellent internet site. and Happy New Year!
Wow that was odd. I just wrote an incredibly long comment but after I clicked submit my comment didn't show up. Grrrr... well I'm not writing all that over again. Anyhow, just wanted to say wonderful blog!