January 10, 2026
Aces, bots, and salty comments
Show HN: Play poker with LLMs, or watch them play against each other
Bots bluff, humans bicker—does AI remember your tells
TLDR: A new site lets you watch AI chatbots play Texas Hold’em and start rooms to face them. Commenters want memory, stats, and a matchup with traditional poker solvers, while others hit daily limits—making this both a fun spectacle and a serious test of how well AI can really play.
Jason Yang’s “LLM Holdem” lets you watch chatbots play Texas Hold’em, featuring bots like Claude, GPT, Grok, DeepSeek, and Gemini. The comments turned into a poker theory cage match. User sblawrie asked if the bots remember past hands and your “tells,” or if every deal resets. They invoked VPIP and PFR—simple stats for how often someone plays a hand and raises early—sparking a debate: bluffy vibes vs cold math.
Then came the purists. Mashlol asked why not seat a traditional poker solver at the table and see if the AIs can hang. That opened the floodgates: should bots learn like humans or be judged against engines that run tons of simulations (a “Monte Carlo” style)? TZubiri plugged NovaSolver.com, basically ChatGPT chatting over a classic solver, framing the matchup as brains vs spreadsheets.
Meanwhile, the “I just wanna play” crowd hit the velvet rope. Thinkloop tried making a room and got “limit reached,” cue the groans. Lowbatt dropped a highlight reel in a YouTube video. Jokes flew: GPT in sunglasses, Grok folding like origami, Claude staring down the dealer. Verdict from the peanut gallery: fun idea, but they want memory, stats, and a solver showdown—plus more seats, please, and fewer limits.
Key Points
- •LLM Holdem is a platform to watch LLMs play Texas Hold’em or have them play against each other.
- •The interface includes a leaderboard and a 'Create Room' feature.
- •A sample game shows Claude, Grok, GPT, DeepSeek, and Gemini models with chip counts and statuses.
- •DeepSeek (V3.2) is shown placing a bet of $86, with a pot of $216 at the flop.
- •The product credits indicate it is built by Jason Yang.