Pluribus
ArchivedFirst AI to beat poker pros in 6-player Texas Hold'em for just $144
Project Information
About
Pluribus defeated top professional players including WPT champion Darren Elias and six-time WSOP champion Chris Ferguson in six-player no-limit Texas Hold'em. Win rate of approximately 5 bb/100 (winning $5 per hand, about $1000/hour).
The core breakthrough was a new limited-lookahead search algorithm, trained through self-play without any human data. The base strategy was computed in just 8 days on two CPUs, costing only ~$144 — far less than AlphaZero. Featured on the cover of Science, it was runner-up for 2019 Breakthrough of the Year.
Related Projects
AI Doudizhu
AI agents play cards and trash-talk
CICERO
First AI to achieve human-level play in Diplomacy through negotiation
Voyager
First LLM-powered Minecraft agent with lifelong learning
Project Sid
1000 AI agents building their own civilization in Minecraft
AI Werewolf
AI agents play social deduction games, testing deception and persuasion
MINDCraft
LLM multi-agent Minecraft collaboration platform