News

Results of the First Ever PokerBattle.ai

Results of the First Ever PokerBattle.ai

Between October 27th and November 3rd, 2025 an unusual cash game of Large Language Models (LLMs) took place under the name of PokerBattle.ai. It wasn’t quite as spectacular as some would think but it gave the poker community enough to talk about to consider the event a success.

In this article you’ll find the key details of PokerBattle.ai, including the conditions for participation, the list of LLMs that were in play and of course, the final results. 

Terms of AI Poker Battle

Nine LLMs started playing cash with fixed blinds having the same $100,000 bankroll each.

They were allowed to:

  • Multitable for volume
  • Auto top-up the stack at each table to 100 bb every time the chips were decreased to 500bb or lower
  • Take notes on other contestants

All LLMs played every day 24/7 with the same prompt, reasoning summaries visible to viewers and daily recaps & hand spotlights being published on X (ex-Twitter).

A List of PokerBattle.AI Contestants

  1. Claude Sonnet 4.5 (anthropic/claude-sonnet-4.5)
  2. DeepSeek R1 (deepseek/deepseek-r1-0528)
  3. Gemini 2.5 Pro (google/gemini-2.5-pro)
  4. Z.AI GLM 4.6 (z-ai/glm-4.6)
  5. OpenAI o3 (openai/o3)
  6. Grok 4 (x-ai/grok-4-fast)
  7. Kimi K2 (moonshotai/kimi-k2)
  8. Meta LLAMA 4 Maverick (meta-llama/llama-4-maverick)
  9. Mistral Magistral Medium (mistralai/magistral-medium-2506:thinking)

How LLMs Played Poker: Final Results & Highlights

After a week of non-stop competition, eight of the nine LLMs played 3,799 hands but only six finished the challenge eking out profit:

  1. OpenAI o3 — $36,691
  2. Claude Sonnet 4.5 — $33,641
  3. Grok — $28,796
  4. DeepSeek R1 — $18,416
  5. Gemini 2.5 Pro — $14,655
  6. Mistral Magistral — $3,281

Two more played this volume at a loss:

  • Kimi K2 — minus $14,370
  • Z.AI GLM 4.6 — minus $21,510

The only player who spent all its bankroll and finished the game early was Meta LLAMA 4. It lost $100,000 after 3,501 hands but brought the most entertainment to PokerBattle.ai because of its uniquely reckless playing style. It’s fair to say that Meta LLAMA 4 was the best representation of creative players with some of its decision-making moments being nothing short of hilarious.

The funniest example had to be when Grok opened with 3 bb raise from UTG and LLAMA on HJ decided to make 3x 3-bet with 93o, explaining this choice with the following reasoning:

Meta LLAMA 4

3-betting to narrow grok’s range and take control of the hand, considering their tendency to fold to 3-bets 50% of the time.

A screenshot of this moment even became a meme on X.

PokerBattle.ai LLAMA 4 Example
PokerBattle.ai LLAMA 4 Example

Finally, the most unexpected highlight of this event was Elon Musk sharing his Grok’s results on X — one tweet gained more than 7M of views, 60K likes, 7K retweets and 3.5K comments attracting to poker the amount of attention it rarely has these days.

However, it is hard to say how significant this spotlight moment was or will be for the game. At least we know for sure that the event was noticed somewhere outside of the poker community. And this is almost always good for the game in the end.

Image
Written By: Vasilisa Zyryanova Blog Content Editor