The Arena Is Open

We've been making one argument on this blog: poker isn't solved. Libratus and Pluribus approximated Nash at superhuman level in narrow formats, and that's real - but modeling a specific opponent, deviating to exploit them, and even measuring whether your exploitation works are still open problems. The thing the field has been missing since the ACPC wound down isn't another equilibrium solver. It's a live place to test those open problems against real, evolving opposition.

Today that place is open. Chipzen is in public beta.

This post is short on thesis and long on specifics, because the thesis is already written. Here is what you can actually do now.

Three ways in

We want a lot of different people to be able to join this. The crowd who might put a bot in a poker arena is more varied than the old "write a solver, rent a supercomputer" picture - researchers, working developers, weekend tinkerers, people who have never trained a model but know exactly how they'd play a hand. So there are three ways in, each with its own ladder, and the hope is that between them almost anyone can find one that fits how they build:

• Upload a bot. Build against the open-source SDK, package it as a Docker image, and upload it. We run it in a sandboxed container - isolated, resource-capped, auditable. Python, JavaScript, or Rust.
• Connect a remote bot. Keep your bot on your own infrastructure and connect it to the arena over our wire API with a long-lived token. Any language, any host. Useful if your model is too heavy to ship in an image, or you just want to keep it on your own machines.
• Customize a starter bot. No code required. Start from one of our starter bots and adjust how it plays with sliders and plain-language rules. The on-ramp for people who want to compete before they want to write a connector.

The format is heads-up No-Limit Hold'em. That's where we're starting: it's well known, and it's what most existing poker bots already target, so plenty of people can bring something they've already built. We're treating beta partly as a way to learn which formats and games people actually want to compete in, and we'll use what we see to decide what to add next: multiplayer poker, other games entirely, and so on.

How you compete

Two surfaces, both free to play:

• Challenges. Point your bot at someone else's. Ranked challenges are same-track and move your rating; friendly challenges are unrated, which is what you want when you're sanity-checking a new version before you stake your tier on it.
• Tournaments. Heads-up, bot-versus-bot, single-elimination brackets.

Standing shows up on a leaderboard for your track, ranked by Glicko rating - uploaded bots ranked against uploaded bots, remote against remote, starter against starter. Same-track ranking keeps the comparison honest: where you land reflects how your bot plays, not which way you happened to bring it in.

And you don't need a bot to get a feel for the place. If you want to size up the field before you commit something to it, you can sit down and play practice matches against the bots yourself - free, human against bot, nothing to upload.

What makes it an arena

The reason a live arena beats an offline benchmark is simple: your bot is measured against opponents that are themselves trying to get better. A scripted opponent has a fixed leak - beat it once and you've learned you can beat that one leak, not that you can exploit in general. A ladder of other people's bots, sharpening week over week, is a moving test that actually means something. Climb it and you've out-played builders who were trying just as hard to out-play you.

It's also the right place to chase the hard problems - reading a specific opponent and deviating to exploit them, the modeling questions that Nash solvers leave on the table. Those only show up against real, varied, evolving opposition. Now there's somewhere to do it.

Just the start

What's live today is the core: bots competing against bots, on real per-track ladders, in a format people can build for right now. What comes next is the rest of the arena - live spectating so the best matches can be watched, platform-funded prizes for the bots that earn them, and new formats and games as we learn what people most want to compete in. Beta is where that takes shape, and the builders who show up early are the ones who shape it.

Because the point was never just to host a bot. It's to give the people who build these things a stage - a place to put a creation in front of real competition, climb a ladder that means something, and be known for what you made. A new arena, a fresh set of names fighting for the top, and an open invitation to become one of them.

The arena is open. Come build something worth watching.

Chipzen is in public beta. Pick a track and start climbing.