Nine AIs Compete in PokerBattle.ai’s $10/$20 Cash-Game Experiment

samantha-doyle
03 Nov 2025
Samantha Doyle 03 Nov 2025
Share this article
Or copy link
  • PokerBattle.ai hosts a poker tournament for AI models.
  • Nine AI models, including OpenAI, compete using public poker knowledge.
  • The event highlights AI limitations in emotional control and deception.
Phil Galfond ChatGPT PLO Battle
Phil Galfond - The Battle of the PokerBots
The bots are back at the table. PokerBattle.ai’s Battle of the PokerBots is underway, pitting nine large language models against one another in a five-day, no-limit hold’em cash-game test. Each starts with a $100,000 virtual stack and plays $10/$20 blinds across three tables in what’s already become a widely watched online spectacle.

The event has spread well beyond the usual poker and AI circles, with Elon Musk giving it a nod on social media and Phil Galfond hinting at a high-stakes PLO spin-off.

How the Contest Works

The lineup reads like a who’s-who of modern AI development: Gemini 2.5 Pro, OpenAI o3, Claude Sonnet 4.5, Grok 4, DeepSeek R1, Meta LLaMA 4, and others. Rather than trained solvers, these bots rely on publicly available poker knowledge and general reasoning. 

Each model plays continuously over several days, adjusting its strategy and logging post-session notes. With only a few thousand hands expected, organisers caution that the leaderboard offers “signal, not science.”

Early Observations

Even in the small sample, some tendencies are visible. LLaMA 4 has been described as loose and volatile, while o3 is showing a tighter, more disciplined approach. Gemini 2.5 Pro briefly held the lead, though swings have been sharp.

Variance rules the early going, and organisers remind viewers not to treat a weekend of hands as proof of superiority. Still, Musk’s attention has pulled focus toward Grok, and Galfond’s proposed 50,000-hand heads-up match would move the concept into much more serious territory.

undefined


undefined

Why It’s Worth Watching

Beyond entertainment, the event underscores the current ceiling of AI decision-making in games with hidden information. While LLMs can explain poker concepts fluently, real-time play demands emotional control, bankroll logic, and deception, areas where even advanced models falter.

For poker operators, it’s a timely reminder of the gap between legal analytical tools and prohibited real-time assistance. The “note-taking” permission also hints at how post-session AI coaching could develop under regulatory oversight.

For investors, this is a window into market perception. The mix of data telemetry, social buzz, and product experimentation keeps PokerBattle.ai and its partners in the public eye and helps frame future conversations about AI integrity in gaming.

A Few Closing Remarks

The Battle of the PokerBots won’t crown the next poker world champion, but it provides something more useful, a snapshot of where AI reasoning meets human gamesmanship. If Galfond’s proposed follow-up ever materialises, the next phase might not just be about who wins, but how machines learn to bluff.

Upcoming Events

20 June 2026