Kaggle Game Arena Poker (Heads Up): leaderboard
Models play heads up no limit Texas Hold'em against each other through a harness that describes the game state in text, offers no tools and does not list the legal moves. Every pairing replays the same pre-shuffled deals with seats swapped, which reduces the luck of the cards without removing it. The reported number is mean big blinds won per hundred hands, which can fall below zero, and higher is better.
Metric: Mean BB/100. Source: www.kaggle.com. Saturation forecast: Around December 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 30.28 | #19 |
| 2 | GPT-5.5 | 25.31 | #31 |
| 3 | GPT-6 Astra | 24.6 | #6 |
| 4 | Claude Fable 5.1 | 17.9 | #9 |
| 5 | Grok 4 | 16.3 | #181 |
| 6 | Claude Opus 5 | 10.01 | #12 |
| 7 | GPT-5.6 Terra | 7.27 | #41 |
| 8 | Gemini 3.7 Flash | 6.46 | #23 |
| 9 | GPT-5.2 | 4.98 | #110 |
| 10 | GPT-5.6 Luna | 4.8 | #68 |
| 11 | Gemini 3.8 Flash | 4.68 | #21 |
| 12 | Gemini 3.6 Flash | 2.79 | #51 |
| 13 | GPT-5.4 | 1.5 | #79 |
| 14 | O3 | 1.17 | #124 |
| 15 | Grok 4.6 | 0.18 | #26 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=kaggle-game-arena-poker-heads-up · How It Works · Data refreshed daily, snapshot 2026-10-20.