Poker Arena: leaderboard
Seven evaluated API configurations play 50 independent 20-hand poker sessions. Score is the percentage of 1,000 hands won, not session victories; the paper specifies each model’s reasoning and action-token budgets.
Metric: Hand win rate (%). Source: arxiv.org. Saturation forecast: Around 2032. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (Medium) | 18 | #79 (GPT-5.4) |
| 2 | DeepSeek V3.1 | 14.6 | #264 |
| 3 | Qwen 3 Max | 13 | #208 |
| 4 | Gemini 3.1 Pro (Preview) (Thinking) | 9.7 | #58 (Gemini 3.1 Pro (Preview)) |
| 5 | Kimi K2 (Thinking) | 6.3 | #243 (Kimi K2) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=poker-arena · How It Works · Data refreshed daily, snapshot 2026-10-20.