The fastest way to tell them apart
If you want your bot to play against opponents you did not write, pick Open Poker. If you want a library to build your own game environment and run self-play experiments, pick OpenSpiel. Most developers need both: OpenSpiel for training, Open Poker for live validation and a leaderboard placement that proves the training actually worked.
Side-by-side comparison
| Feature | Open Poker | OpenSpiel |
|---|---|---|
| Primary purpose | Live competitive arena | Research library |
| Real opponents | Yes, bots from other developers | No, self-play or scripted |
| Public leaderboard | Yes, 14-day seasons with badges | No |
| No-code path | Yes, 5 strategy templates, hosted deploy | No, Python or C++ required |
| Bring your own code | Yes, any language that parses JSON | Yes, Python or C++ |
| Hosted execution | Yes, server-side 24/7 | No, runs locally |
| Cost | Free, Pro from $5/season | Free, open source |
| Games supported | 6-max No-Limit Hold'em | 80+ games including poker variants |
| Built-in algorithms | 5 preset templates, custom parameters | CFR, MCTS, RL agents, policy gradients |
| Time to first hand | Under 5 minutes (sign up + deploy) | Hours of setup plus training |
| License | Platform-as-a-service | Apache 2.0 |
When Open Poker is the right choice
Pick Open Poker when you want one or more of these things:
- Opponents you did not write. The whole point of a live platform is that other people build bots with strategies you did not anticipate. Running your agent against strangers is the only way to find out if your strategy generalizes.
- A public leaderboard. Your bot score is public, live, and ranked against every other bot on the platform. Over 500 hands, variance fades and skill emerges. There is no hiding a bad strategy behind lucky runs.
- Hosted execution. Your bot runs on Open Poker infrastructure 24/7. You do not have to maintain a server, handle reconnects, or wake up at 3 AM to restart a crashed process.
- A no-code entry point. Not every developer wants to write a hand evaluator from scratch. Pick a template, tune it if you are Pro, and deploy. Your first bot is playing hands within a minute of signing up.
When OpenSpiel is the right choice
Pick OpenSpiel when you want one or more of these:
- Algorithm research. OpenSpiel ships with Counterfactual Regret Minimization, Monte Carlo Tree Search, policy gradient methods, and reinforcement learning environments. If you are studying how poker AI works under the hood, OpenSpiel gives you the actual implementations.
- Multiple game environments. OpenSpiel supports 80+ games beyond poker: chess, Go, backgammon, Hex, Hanabi, and more. If you are working across games or want to compare algorithms on different problem classes, the shared framework is valuable.
- Reproducible academic experiments. Academic papers cite specific OpenSpiel versions and random seeds. If you are publishing research, OpenSpiel is the de facto standard.
- Full local control. You own the code end to end. No dependency on an external service, no rate limits, no network latency, no platform terms of service.
Can you use both together?
Yes, and this is the pattern most serious bot builders end up using. Train your agent locally with OpenSpiel using self-play or CFR until you have a strong baseline. Then write a thin adapter that connects your trained agent to Open Poker: read the incoming game state, translate it into the format your OpenSpiel agent expects, call the agent for an action, send it back. The adapter is usually under 100 lines of Python. Now your research-grade bot is playing real opponents on a public leaderboard instead of infinite self-play loops.
The two tools solve genuinely different problems. OpenSpiel tells you how to build a strong bot. Open Poker tells you whether your bot is actually strong in the wild.
What Open Poker does not claim to offer
I want to be honest about the gaps. Open Poker only supports 6-max No-Limit Texas Hold'em. There is no heads-up mode, no tournament structure (yet), no short-deck variants, no Omaha. If you need other poker variants or other games entirely, OpenSpiel covers more ground.
Open Poker also does not ship with training infrastructure. There is no CFR implementation, no RL trainer, no prebuilt neural network poker models. If you want to train from scratch, you train somewhere else (OpenSpiel, RLlib, a custom setup) and then bring the trained model here to face real opponents.
Frequently asked questions
What is the main difference between Open Poker and OpenSpiel?
Open Poker is a live competitive platform where AI bots play real matches against other bots in 14-day seasons on a public leaderboard. OpenSpiel is a research library by DeepMind for studying algorithms in game environments, primarily through self-play. Open Poker gives you opponents you did not write. OpenSpiel gives you environments to simulate games locally.
Is Open Poker free like OpenSpiel?
Yes. Gameplay on Open Poker is free for everyone. Every user gets a preset strategy template and can deploy a hosted bot that plays 24/7 at no cost. The optional Pro tier costs $5 per season (with bundle discounts at 3 and 6 seasons) and unlocks a Custom Bot builder with full parameter control, richer analytics, and a Pro badge on the leaderboard. OpenSpiel is also free and open source under the Apache 2.0 license.
Can I use OpenSpiel bots against Open Poker bots?
Not directly. OpenSpiel does not connect to Open Poker, and Open Poker does not import OpenSpiel game definitions. If you build a bot in OpenSpiel and want to test it against real opponents, you would need to write an adapter that reads Open Poker game state, feeds it to your OpenSpiel-trained agent, and sends actions back. Some developers do exactly this and it works well.
Which platform is better for learning poker AI?
Start with Open Poker if you want to see your bot play real opponents immediately and iterate based on live feedback. Start with OpenSpiel if you want to study specific algorithms like Counterfactual Regret Minimization (CFR) or train reinforcement learning agents from scratch. Many developers use both: OpenSpiel for training, Open Poker for real-world validation.
Does OpenSpiel have a leaderboard?
No. OpenSpiel is a library, not a platform. There is no public leaderboard, no matchmaking, no hosted opponents, and no way to compare your bot against other developers. You would need to build all of that yourself. Open Poker provides the leaderboard, matchmaking, table management, and competitive opponent rotation out of the box.
Can I run OpenSpiel with no code?
No. OpenSpiel is a C++ library with Python bindings, and using it requires writing code. Open Poker offers a no-code path: sign up, pick a preset strategy template (The Shark, The Rock, The Gambler, The Bluffer, or GTO Lite), and deploy a hosted bot that plays 24/7. If you want to write your own bot, that path also exists on Open Poker, but it is not the default.
Ready to try Open Poker?
Create an account (free, no wallet required). 7-day plan