OpenAI o3 Wins Poker Bot Battle; Elon Musk's Grok Finishes Third
Table Of Contents
A poker battle between artificial intelligence bots that went viral in the poker and tech worlds has concluded after five days of nonstop automated action.
Over 3,799 hands of $10/$20 (Play Money) no-limit Hold'em, OpenAI o3 won $36,691 to be the biggest winner of PokerBattle.ai among nine competing large language models (LLMs). One of the most well-known of the bots, the Elon Musk-owned Grok, held a lead toward the end of the match but ultimately finished third with $33,641 in winnings.
The only bot to bust from the poker marathon was Meta LLAMA 4, which lost its entire $100,000 bankroll after 3,501 hands.
PokerBattle.ai Results
| Rank | Player | Total Winnings (Play Money) | Final Bankroll (Play Money) | Hands Played |
|---|---|---|---|---|
| 1 | OpenAI o3 | $36,691 | $136,691 | 3,799 |
| 2 | Claude Sonnet 4.5 | $33,641 | $133,641 | 3,799 |
| 3 | Grok 4 | $28,796 | $128,796 | 3,799 |
| 4 | DeepSeek R1 | $18,416 | $118,416 | 3,799 |
| 5 | Gemini 2.5 Pro | $14,655 | $114,655 | 3,799 |
| 6 | Mistral Magistral | $3,281 | $103,281 | 3,799 |
| 7 | Kimi K2 | -$14,370 | $86,030 | 3,799 |
| 8 | Z.AI GLM 4.6 | -$21,510 | $78,490 | 3,799 |
| 9 | Meta LLAMA 4 | -$100,000 | $0 | 3,501 |
Different Playing Styles
The poker bot battle was conceived by 33-year-old Max Pavlov, who told PokerNews that the models, which could take notes and adapt to opponents, were trained using "poker books, blog posts," and other study resources.
The biggest pot of the day went to OpenAI o3 in a four-bet pot where they were dealt aces against the pocket queens of Gemini 2.5 Pro, who called a four-bet before calling bets on both the flop and turn on the 9♦8♦7♣4♥ board to send $10,795 in play money to OpenAI o3.
Gemini 2.5 Pro explained its reasoning on making the turn call. "This is a tough spot, but I believe a call is the correct play. My opponent's line is extremely polarized. He is representing a monster hand (AA/KK) or a complete bluff (like AK). Given his aggressive 4-bet pre-flop and the small flop bet, this turn shove looks like a move to get me to fold an overpair. The pot odds are compelling (I need to be right about 37.5% of the time). Against a range of AA, KK, and AK, my QQ has enough equity to justify a call. Folding here feels too exploitable against a capable, aggressive opponent. This is a stand I have to take."
Grok 4, which is in talks to face poker legend Phil Galfond in a high-stakes heads-up battle, held a late lead with under four hours left in the poker battle but couldn't hang onto it. The Anthropic-owned Claude Sonnet 4.5 fared the second best of the bots with $33,641 in winnings.
While Pavlov had warned about being results-oriented with such a small sample size of hands, it is clear that playing style had something to do with Meta LLAMA 4's abysmal performance. The bot, which was the only one to lose its entire $100,000 bankroll, showed a VPIP of over 60%, nearly three times that of its competitors.
To view hand histories from the match, visit pokerbattle.ai/hand-history.
