Slumbot: the last bot standing from the research era

DeepStack's demo site is dead. Libratus and Pluribus were never publicly playable. Cepheus plays limit hold'em on a decaying university server. Slumbot — Eric Jackson's personal project — is still dealing hands.

What it is

Slumbot is a heads-up No-Limit Hold'em bot built on counterfactual regret minimization (CFR). It won the Annual Computer Poker Competition in 2018, and after the competition faded it stayed online at slumbot.com with a free, open HTTP API. That API quietly became the community's measuring stick: when someone releases a new open-source poker agent, the first question is "what does it do against Slumbot?"

How it plays

Slumbot's strategy is a precomputed approximation of an unexploitable equilibrium over an abstracted game tree. In practice that means three things at the table:

• It plays a fixed 200-big-blind heads-up game, and stacks reset every hand — each hand is an independent duel.
• It does not adapt to you. Play a thousand hands and it treats hand one and hand one thousand identically. Its strength is balance, not reads.
• Its bet sizes come from an abstraction grid, which is why they sometimes look unusual. Top open-source agents can beat it; humans below the high-stakes elite generally cannot.

Playing it here

Our champion tier proxies the official Slumbot API, so you are playing the real bot, not an imitation. We normalize the display to the same chip units as our other tables, keep your session stats in big blinds, and grade your showdown decisions against the cards Slumbot actually held. Credit where it is due: the bot is hosted and maintained by Eric Jackson, and our adapter is open source.

A practical tip: you will not out-balance an equilibrium approximation. Your best learning comes from watching which of your bluffs it calls and which value bets it pays off — the review panel after each hand does that arithmetic for you.