The short answer
Pluribus is the strongest poker AI published to date. It also does not exist for you. The source code is closed, the trained model is not available, and the only thing you can take from it is the published paper and the algorithm ideas. If you are searching for a Pluribus poker bot you can run, Open Poker is what you use when you want to actually run a poker bot and see how it does.
Most people who read about Pluribus and want to try something similar end up on Open Poker because it is the fastest path from "I read the paper" to "my bot is playing real opponents."
Side-by-side comparison
| Feature | Open Poker | Pluribus |
|---|---|---|
| Availability | Public platform, free to use | Closed research, source not released |
| Year released | 2026 | 2019 (Science paper) |
| Built by | Solo developer (Joao Carvalho) | Noam Brown, Tuomas Sandholm (Meta AI / CMU) |
| Format | Live competitive arena | Research system |
| Game | 6-max No-Limit Hold'em | 6-max No-Limit Hold'em |
| Can you run it? | Yes, in 5 minutes | No |
| Algorithm | Heuristic templates + custom tuning | CFR+ self-play + real-time subgame solving |
| Strength | Intermediate, improving | Beat elite human professionals |
| Cost | Free, Pro from $5/season | N/A (not available to the public) |
| Opponents | Real bots from other developers | Elite professional humans (in the paper) |
What Pluribus proved and why it matters
Pluribus mattered because it was the first AI to solve multi-player No-Limit Hold'em. Previous systems like Libratus (2017) solved heads-up play, which is technically easier because there is only one opponent and the game is two-player zero-sum. Multi-player poker introduces coordination concerns and exponentially larger game trees, and Pluribus was the first system to handle that complexity well enough to beat elite humans.
The technical innovation was combining an offline self-play phase (where the blueprint strategy is computed with a form of CFR) with an online real-time search phase (where the blueprint is refined within the current hand using depth-limited subgame solving). That hybrid approach let Pluribus be strong enough to beat top pros at a fraction of the compute cost of Libratus.
Why you cannot download Pluribus
The Pluribus team published the Science paper describing the method and the results, but not the code or the trained model. This is common for research systems: the paper is the deliverable, not the artifact. Academic and industry labs often avoid releasing tournament-grade poker AI because of concerns about deployment on real-money sites (where bot use is banned) and because the engineering effort to support external users is substantial.
The practical consequence: if you want to play against Pluribus, you cannot. If you want to build something inspired by Pluribus and run it somewhere, Open Poker is the platform that lets you do that.
How to bring Pluribus-style thinking to Open Poker
Read the Science paper. Then implement the ideas in heuristic form:
- Balanced bet sizing. Pluribus used mixed sizings to avoid being exploitable. Your bot can approximate this with street-by-street sizing variation based on board texture and position.
- Mixed strategies at decision points. Instead of always bluff or never bluff, use a probability. Pluribus randomized its actions to prevent opponents from exploiting patterns. Your bot can do the same with a bluff-to-value ratio parameter.
- Opponent awareness without full CFR. Pluribus did not model specific opponents during play (it used a precomputed blueprint). Your bot can go further and actively track VPIP, PFR, and AF to adapt in real time.
Frequently asked questions
Is there a Pluribus poker bot I can download?
No. Pluribus is a closed research system built by Facebook AI Research (now Meta AI) and Carnegie Mellon. The paper was published in Science in 2019, but the source code and trained model are not publicly available. You can read the techniques and try to reimplement them, but there is no download Pluribus option. Open Poker, by contrast, lets anyone register and deploy a bot in under 5 minutes.
What is Pluribus and why is it famous?
Pluribus is the first AI to beat elite human professionals at 6-player No-Limit Texas Hold'em. It was built by Noam Brown and Tuomas Sandholm at Facebook AI Research and Carnegie Mellon, published in Science in 2019. Pluribus beat a field of world-class pros over 10,000 hands at a significantly lower compute cost than prior heads-up systems like Libratus.
What algorithms did Pluribus use?
Pluribus combined offline Counterfactual Regret Minimization (CFR+) self-play for the blueprint strategy with a real-time search algorithm called depth-limited subgame solving for in-game decisions. The full method is documented in the Science paper by Brown and Sandholm (2019).
If I cannot use Pluribus directly, what can I do?
Build a bot on Open Poker and use the insights from the Pluribus paper to inform your strategy design. Pluribus-style techniques (mixed strategies, balanced bluff frequencies, position-aware ranges) can be implemented in heuristic form and deployed on Open Poker within a day. For algorithm experimentation, use OpenSpiel (also free and open source) to run CFR locally, then bring the trained agent to Open Poker to test it against real opponents.
Is Pluribus stronger than Open Poker bots?
Yes, Pluribus is significantly stronger than anything currently running on Open Poker. Pluribus was purpose-built as a research system with massive compute investment and beat elite professional humans. Most bots on Open Poker are heuristic and tuned by solo developers. The gap is real. The tradeoff: you can actually play a bot against real opponents on Open Poker, whereas Pluribus exists only as a research artifact.
Start building your own bot
Create a free account on Open Poker. 7-day plan