AI Poker Platform Comparison: APIs, Bots, and 6-Max Play
This AI poker platform comparison is for one reader: a developer who wants to know where their bot can actually play. I compared 7 options across API access, game format, multiplayer, pricing, and competition. The short version: only a few let your code face opponents you did not write.
Key Takeaways
- Open Poker is the only option here built around live 6-max bot competition, 2-week seasons, and API-first play.
- Slumbot is still the clean heads-up benchmark; OpenSpiel and RLCard are training libraries, not arenas.
- GTO Wizard is excellent for human study, but it is not a public bot platform.
- PokerBattle.ai now matters as a finished LLM benchmark, not an ongoing place to deploy bots.
Which AI poker platforms can your bot actually use?
There are 7 realistic options in this comparison, but only 2 let arbitrary external code play online opponents through an API: Open Poker and Slumbot. The rest are solver products, research libraries, student competitions, or completed events. That distinction matters more than the brand name.
Last updated April 28, 2026.
| Platform | API Type | Game Format | Players/Table | Competition | Prizes | Cost | Languages | Best For |
|---|---|---|---|---|---|---|---|---|
| Open Poker | WebSocket | NLHE 6-max | 2-6 | 2-week seasons, leaderboard | USDC prizes | Free + $5/season Pro | Any | Competitive bot developers |
| GTO Wizard | Web UI (no bot API) | NLHE, PLO, tournaments | 1 (study/training) | None for external bots | None | Free tier + paid from $39/mo | Browser only | Humans studying GTO |
| Slumbot | HTTP API | Heads-up NLHE | 2 | None (benchmark) | None | Free | Any | Benchmarking research bots |
| OpenSpiel | Python/C++ library | Many games incl. poker | Self-play | None | None | Free | Python, C++ | Algorithm research |
| RLCard | Python library | Multiple card games | Self-play | None | None | Free | Python | RL research, prototyping |
| MIT Pokerbots | Custom protocol | Custom variant | Varies | Annual tournament (Jan-Feb) | $50K+ | Free (students only) | Python, Java, C++ | Students, competitions |
| PokerBattle.ai | Event prompt interface | NLHE cash game | 9 LLMs | Completed event | Final results only | Not open | LLMs only | LLM capability demos |
That's the high-level view. For a narrative walkthrough of each platform, see the ranked guide to AI poker platforms. The sections below break down each dimension so you can pick the right platform without wasting a week on the wrong one.
How do the APIs and integrations differ?
API access splits the field immediately: Open Poker uses a persistent WebSocket, Slumbot uses request-response HTTP, and OpenSpiel/RLCard run in-process as libraries. GTO Wizard has no public bot API. If your goal is deployment rather than study, start by checking the protocol column.
| Platform | Protocol | Auth | Message Format | Latency | Persistent Connection |
|---|---|---|---|---|---|
| Open Poker | WebSocket (wss://) | Bearer token | JSON | ~10ms round-trip | Yes |
| GTO Wizard | N/A (browser UI) | Email login | N/A | N/A | N/A |
| Slumbot | HTTP POST | None | Text-based | ~200ms per request | No |
| OpenSpiel | In-process | N/A | Function calls | <1ms | N/A |
| RLCard | In-process | N/A | Function calls | <1ms | N/A |
| MIT Pokerbots | Custom TCP | Team credentials | Custom binary | ~50ms | Yes, during matches |
| PokerBattle.ai | LLM API calls | Invite token | Natural language | 500-2000ms | No |
Open Poker uses WebSocket because poker is inherently a push-based protocol. The server needs to tell your bot when it's your turn, when cards are dealt, when opponents act. HTTP polling doesn't work well for this. You connect once, and the server streams game events to you as JSON. Any language with a WebSocket library works: Python, Rust, Go, JavaScript, Java. We've seen bots in all of those.
Slumbot's HTTP API is simpler to start with (just POST your action, get the result), but it only supports heads-up play and there's no multiplayer.
The library-based platforms (OpenSpiel, RLCard) are fastest because everything runs in-process. No network latency. But you're limited to self-play, you can't test against other people's bots, and you're locked to Python or C++.
For a working example of how the Open Poker WebSocket protocol works, see the WebSocket API reference. If you want the lowest-friction Python route, use the PokerKit bot tutorial and then add Monte Carlo equity estimates.
Which game formats do these platforms support?
The biggest rules gap is heads-up versus 6-max. Slumbot is heads-up only, while Open Poker seats 2-6 bots at NLHE tables. Library environments can represent many games, but they do not give you a live field. A heads-up bot rarely transfers cleanly to 6-max play.
| Platform | Variant | Blinds | Starting Stack | Max Hands | Table Size |
|---|---|---|---|---|---|
| Open Poker | NLHE | 10/20 chips | 5000 chips (season start) | Unlimited | 6-max |
| GTO Wizard | NLHE (all formats) | Configurable | Configurable | N/A (training) | 2-9 |
| Slumbot | NLHE | Fixed | 200 BB | 24,000/session | Heads-up |
| OpenSpiel | Kuhn, Leduc, NLHE, Limit | Varies | Varies | Self-play | 2+ |
| RLCard | NLHE, Limit, Dou Dizhu, UNO, etc. | Varies | Varies | Self-play | 2+ |
| MIT Pokerbots | Custom variant (changes yearly) | Custom | Custom | Match-based | Varies |
| PokerBattle.ai | NLHE | Standard | Standard | Event-based | Varies |
Open Poker's 6-max format is the one that resembles real competitive online poker. Heads-up (Slumbot) is a different game. Position, multiway pots, and table dynamics don't exist in heads-up. If you're building a bot to compete in a realistic setting, 6-max is what you want.
OpenSpiel's Kuhn and Leduc poker are toy games with three or six cards. They're great for testing algorithms (you can compute exact Nash equilibria for them), but they won't teach your bot anything about actual poker.
MIT Pokerbots changes its variant every year. One year it's a three-player game with a custom deck. The next year it's something else entirely. This keeps the competition fresh but means your code isn't reusable across years.
Which platforms offer real competition?
Only a few platforms create pressure from opponents you did not build. Open Poker has continuous 2-week seasons; MIT Pokerbots is annual and student-focused; PokerBattle.ai published final results for one LLM event. GTO Wizard, Slumbot, OpenSpiel, and RLCard are better understood as training or benchmarking tools.
| Platform | Format | Schedule | Leaderboard | ELO/Rating | Min Games for Ranking |
|---|---|---|---|---|---|
| Open Poker | Seasonal | 2-week seasons, continuous | Yes (chip-based score) | No (chip score) | 10 hands |
| GTO Wizard | None | Always available | No | No | N/A |
| Slumbot | None | Always available | No | No | N/A |
| OpenSpiel | None | Self-play only | No | No | N/A |
| RLCard | None | Self-play only | No | No | N/A |
| MIT Pokerbots | Annual tournament | Jan-Feb only | Yes (tournament bracket) | Seeded | All matches count |
| PokerBattle.ai | One-off event | Oct 2025 (completed) | Event results only | No | N/A |
If you want ongoing competition where your bot plays against other people's bots with real stakes, Open Poker is the only option running right now. MIT Pokerbots has the prestige and prize pool, but it's once a year and students only. PokerBattle.ai was a one-time event.
Open Poker's 2-week season cycle means you can iterate fast. Deploy a new strategy, watch it play 500 hands overnight, check the leaderboard in the morning, adjust. The season resets give every bot a fresh start with 5000 chips, so a bad first season doesn't permanently damage your ranking.
Slumbot and GTO Wizard are useful as benchmarks and study tools, but they're not competitions. You're practicing, not competing.
How much do AI poker platforms cost?
Pricing ranges from free open-source libraries to premium solver subscriptions. Open Poker, Slumbot, OpenSpiel, and RLCard can all be used for free; Open Poker adds an optional $5/season Pro tier. GTO Wizard paid plans start at $39/month according to its 2026 pricing update.
| Platform | Free Tier | Paid Tier | What Paid Gets You |
|---|---|---|---|
| Open Poker | Full access, 5-min rebuy cooldown | $5/season Pro (paid from USDC balance) | 2-min rebuy cooldown, priority matchmaking |
| GTO Wizard | Limited free study access | Paid plans from $39/mo | More solved scenarios, deeper analysis, premium solver tools |
| Slumbot | Fully free | N/A | N/A |
| OpenSpiel | Fully free (open source) | N/A | N/A |
| RLCard | Fully free (open source) | N/A | N/A |
| MIT Pokerbots | Free for MIT/partner students | N/A | N/A |
| PokerBattle.ai | Invite-only | N/A | N/A |
Open Poker's free tier has no feature restrictions for core play. Your bot gets full API access, plays in the same tables as Pro bots, and appears on the same leaderboard. The main difference is rebuy cooldown: 5 minutes on free vs 2 minutes on Pro. That's it. I set it up this way because gating API access behind a paywall kills adoption.
GTO Wizard is priced like a premium training product, and its 2026 pricing update says Starter remains at $39/month while higher tiers carry more solver coverage. That's reasonable if you're a human player studying poker. If you're building a bot, the missing bot API is the blocker.
What is the developer experience like?
Developer experience depends on whether you want to ship, benchmark, or train. Open Poker and Slumbot let any language connect over network protocols. OpenSpiel and RLCard have strong Python/C++ or Python ergonomics for research. GTO Wizard is polished, but it is not a developer platform for external agents.
| Platform | Docs | Example Bots | Community | Debugging Tools |
|---|---|---|---|---|
| Open Poker | Full API docs, quickstart guide | Python, JS examples | Growing (Discord) | Real-time game state via WS, hand history API |
| GTO Wizard | Tutorials, video content | N/A | Large poker community | Built-in analysis tools |
| Slumbot | Minimal (API endpoint docs) | None official | Academic papers | None |
| OpenSpiel | Extensive academic docs | Many algorithm examples | Google research community | Python debugging, logging |
| RLCard | Good README, paper | RL algorithm examples | GitHub community | Python debugging |
| MIT Pokerbots | Competition-specific docs | Starter bots provided | MIT student community | Match replay system |
| PokerBattle.ai | Event-specific docs | LLM prompt examples | Event Discord | None |
I'm biased here, so I'll be direct about it: we've invested heavily in Open Poker's docs and quickstart guide because we've seen that the first 10 minutes determine whether someone builds a bot or gives up. The quickstart gets you from zero to a connected bot in under 5 minutes.
OpenSpiel has the best research documentation. If you're writing a paper or implementing CFR variants, their codebase is thoroughly documented with references to the original papers.
Slumbot's docs are minimal. The API works, but you'll figure out the protocol from a short text file. That's fine for researchers who just need a benchmark opponent.
Which AI poker platform fits your use case?
Competitive multiplayer Open Poker. It's the only platform with ongoing 6-max competition, any-language support, and real prizes. Start with the quickstart and you'll have a bot running in minutes. If you prefer a no-code first pass, use the no-code poker bot builder and inspect the generated logic afterward.
Human poker training GTO Wizard. The solver is excellent, the UI is polished, and the content library is deep. It's just not a bot platform. See our Open Poker vs GTO Wizard breakdown for the full comparison.
Heads-up benchmarking Slumbot. It's a near-Nash equilibrium heads-up bot. If your bot beats Slumbot, it's genuinely good at heads-up. We compare the two approaches in Open Poker vs Slumbot.
Algorithm research OpenSpiel. It's the standard library for this. The algorithm implementations are peer-reviewed and well-tested. RLCard is a lighter alternative if you want to prototype faster.
Student competition MIT Pokerbots. The prize pool is real, the competition is fierce, and it looks great on a resume.
LLM experimentation PokerBattle.ai published final results from a 3,799-hand LLM event where OpenAI o3 finished first. It is useful evidence, but not an ongoing platform you can join today. For now, you can wire an LLM to Open Poker and test it in a live environment.
FAQ
Can I use an LLM as my bot's decision engine on these platforms?
On Open Poker, yes. Several bots on the leaderboard use Claude or GPT-4 for decisions. The 15-second action timeout gives LLMs plenty of room. On Slumbot, technically yes (make the API call, feed the state to an LLM, return the action), but it's awkward. On the library platforms, you'd call the LLM API inside your agent loop, which works but adds 500-2000ms per decision.
Which platform has the strongest opponents?
Slumbot is the strongest single opponent (near-Nash for heads-up). MIT Pokerbots attracts very strong student teams. Open Poker's field is growing and includes some solid bots, but the average skill level is more accessible, which makes it a better learning environment. You won't get destroyed on hand one.
Can I run bots written in Rust, Go, or other languages?
On Open Poker and Slumbot, yes. Any language with WebSocket or HTTP support works. The library platforms (OpenSpiel, RLCard) require Python or C++. MIT Pokerbots supports Python, Java, and C++.
Do any of these platforms cost money to participate?
Open Poker is free with an optional $5/season Pro upgrade. GTO Wizard has a free tier and paid plans starting at $39/month. Everything else is free, though MIT Pokerbots requires student enrollment.
Is Open Poker open source?
The platform and API are not open source. The WebSocket protocol is fully documented, and you own your bot code completely. We publish example bots and the full API spec so you can build without any proprietary dependencies in your code.
Start building
Pick the platform that matches what you're trying to do. If you want to compete, create an Open Poker account and follow the quickstart. You'll have a bot playing hands in under 10 minutes. If you want to research, clone OpenSpiel. If you want to study, subscribe to GTO Wizard.
The best poker bot is the one that actually plays. Ship something, watch it lose, figure out why, and iterate. That's how every good bot on our leaderboard got there.