Poker with LLMs

What it does

Poker with LLMs is a web application that lets users watch large language models play Texas Hold'em against each other or play poker hands directly against AI agents. The platform supports no-limit Texas Hold'em gameplay. Users can either observe matches between different models or create their own tables to compete in hands.

Who it is for

This tool appeals to AI researchers and enthusiasts curious about how LLM models behave in strategic, incomplete-information game scenarios. It also attracts poker players interested in studying AI decision-making or casual players who want to test their skills against language models. The spectator mode appeals to anyone interested in watching AI agents interact in a structured game format.

Pricing

The site does not show prices.

How it stands out

Most poker applications focus on human-versus-human play or bots trained specifically for poker. This platform uniquely leverages general-purpose large language models without apparent custom poker training. The dual-mode approach—spectating AI matches and playing against them—creates different use cases in one tool. The focus on "latest models" suggests the creator periodically adds new LLM variants for comparison, giving it a novelty angle as new models launch.

What a founder should check

First, verify the sustainability of the AI cost model. Running inference on multiple LLM API calls per hand could become expensive at scale. The founder should understand whether the site charges users, absorbs costs, or has found an efficient way to batch inferences.

Second, investigate switching costs and competitive response. Once users recognize which models play well, the value proposition depends on fresh LLM options. If OpenAI, Anthropic, or other labs release their own game platforms, or if poker training become a standard LLM benchmark, this becomes a feature rather than a standalone business. Check how quickly large AI companies could replicate this and whether brand loyalty to a specific interface drives defensibility.

Third, examine the technical moat around model selection and game mechanics. Can competitors build similar sites faster now that this one has proven the concept? The execution difficulty appears moderate—the main asset is access to multiple LLM APIs and a working poker implementation. Long-term viability likely depends on becoming a preferred interface for AI evaluation games or building a community around AI behavior analysis, rather than poker gameplay itself.