Poker AI — Pluribus, Libratus, Cepheus
Last updated: May 16, 2026
Cepheus (2015) solved heads-up Limit Hold'em. Libratus (2017) beat 4 top pros 1-vs-1 NLHE. Pluribus (2019) became the first AI to beat top humans in multi-player (6-max) NLHE. Full multi-player No-Limit Hold'em remains unsolved due to computational complexity. Modern solvers (PioSolver, GTO+) compute approximate GTO for specific scenarios — not full games. AI tools augment human players via study (GTO Wizard, training apps) rather than replacing them.
Definitions
Frequently Asked Questions
Has poker been solved by AI?
Only heads-up Limit Hold'em — solved by Cepheus (University of Alberta, 2015). Heads-up No-Limit Hold'em has been 'beaten' by AI (Libratus 2017) but not fully solved. Multi-player No-Limit (the most common live/online format) is computationally intractable to fully solve. Modern solvers approximate GTO for specific subgames, not full games.
What is Pluribus?
Pluribus is the poker AI built by Facebook AI Research and Carnegie Mellon University. In 2019, it beat 12 top human poker professionals in 6-player No-Limit Hold'em over 10,000 hands. Pluribus uses 'self-play' (playing against copies of itself) plus 'depth-limited search' (looking ahead a few decisions, not entire games). First-ever AI to beat humans in multi-player NL poker.
Can a poker bot beat humans online?
Sophisticated bots can beat low-stakes humans, but major poker sites actively detect and ban them. PokerStars, GGPoker, partypoker employ teams that monitor for bot patterns. Bots typically lose accounts + frozen funds within months. At mid-stakes and above ($1/$2 NL +), human skill exceeds most affordable bots — and operator detection is more aggressive.
What's the difference between solvers and AI?
Solvers (PioSolver, GTO+) compute equilibrium strategies for specific input scenarios (e.g., BTN vs BB single-raised pot on K-7-3 rainbow). AI like Pluribus plays full games end-to-end. Both apply game theory, but solvers are tools used by humans for study; AIs are autonomous players. Solvers ~$75-$475 one-time; AI like Pluribus is research-only (not sold publicly).
Will AI replace human poker players?
Not in the foreseeable future — for several reasons: (1) Multi-player NL is still unsolved; AI advantages are limited to specific structures. (2) Major sites ban bots aggressively. (3) Live poker is human-vs-human; no AI competes. (4) The entertainment value of poker comes from human play. AI tools (solvers, training apps like GTO Wizard) augment human players rather than replace them.
Related
Equity math powered by Monte Carlo
Open RiverOdds Calculator →