OpenAI beats Google, Meta, and Grok in all-AI poker tournament

Date:
Sat, 06 Dec 2025 04:45:00 +0000

Description:
A weeklong AI poker tournament revealed just how strategic and flawed AI models can be when playing under pressure.

FULL STORY ======================================================================OpenAIs o3 model won a five-day poker tournament of nine AI chatbots The o3 model won by playing the most consistent game Most top language models handled poker well, but struggled with bluffing, position, and basic math

In a digital showdown unlike anything ever dealt at the felt, nine of the worlds most powerful large language models spent five days locked in a high-stakes poker match.

OpenAIs o3, Anthropics Claude Sonnet 4.5, X.ai's Grok, Google's Gemini 2.5 Pro, Metas Llama 4, DeepSeek R1, Kimi K2 from Moonshot AI, Magistral from Mistral AI, and Z.AIs GLM 4.6 played thousands of hands of no-limit Texas
hold 'em at $10 and $20 tables with $100,000 bankrolls apiece.

When OpenAIs o3 model walked away from a weeklong poker game $36,691 richer, there was no trophy, just bragging rights.

The experimental PokerBattle.ai was entirely AI-run with the same initial prompt issued to each player. It was pure strategy, if strategy is what you call thousands of micro-decisions made by machines that dont really
understand winning, losing, or how humiliating it is to bust with
seven-deuce.

For a tech stunt, it was unusually telling. The top-performing AIs werent
just bluffing and betting they were adapting, modeling their opponents, and learning in real time how to navigate ambiguity. While they didnt play flawless poker, they came impressively close to mimicking seasoned players' judgment calls.

OpenAIs o3 quickly showed it had the steadiest hand, taking down three of the five biggest pots and sticking close to textbook pre-flop theory. Anthropics Claude and X.coms Grok rounded out the top three with substantial profits of $33,641 and $28,796, respectively.

Meanwhile, Llama lost its full stack and flamed out early. The rest of the pack landed somewhere in between, with Googles Gemini turning a modest profit and Moonshots Kimi K2 hemorrhaging chips down to an $86,030 finish. Gambling AI

Poker has long been one of the best analogs for testing general-purpose AI. Unlike chess or Go, which rely on perfect information, poker demands that players reason under uncertainty. Its a mirror of real-world decision-making in everything from business negotiations to military strategy, and now, apparently, chatbot development.

One consistent takeaway from the tournament was that the bots were often too aggressive. Most favored action-heavy strategies, even in situations where folding would have been wiser. They tried to win big pots more than they
tried to avoid losing them. And they were awful at bluffing, not because they didnt try, but because their bluffs often stemmed from misread hands, not clever deception.

Still, AI tools are getting smarter in ways that go far beyond surface-level smarts. Theyre not just repeating what theyve read; theyre making probabilistic judgments under pressure and learning to read the room. Its
also a reminder that even powerful models still have flaws. Misreading situations, drawing shaky conclusions, and forgetting their own position isnt just a poker problem.

You might never sit across from a language model in a real poker room, but odds are youll interact with one trying to make decisions that matter. This game was just a glimpse of what that could look like.

Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the
Follow button!

And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.



======================================================================
Link to news story: https://www.techradar.com/ai-platforms-assistants/openai-beats-google-meta-and -grok-in-all-ai-poker-tournament


--- Mystic BBS v1.12 A49 (Linux/64)
* Origin: tqwNet Technology News (1337:1/100)