HoldemBench: leaderboard
HoldemBench: duplicate heads-up no-limit Texas hold'em (blinds 50/100, 100 big blinds, stacks reset every hand). Every deal is played twice with seats swapped and all-in hands before the river are scored by exact equity, so most card luck cancels. LLMs, OpenRouter decision models and a local model play through one identical action menu. Score: all-in-adjusted win rate in milli-big-blinds per hand against the whole field, from one additive model over all pairings of six leagues (95% deal-cluster bootstrap intervals upstream). Unbounded; 0 is break-even. Reference bots are excluded. Higher is better. Source: https://huggingface.co/datasets/chrystians/poker-holdembench
Metric: All-in-adjusted win rate (mbb/hand). Source: huggingface.co. Saturation forecast: Around May 2028. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4.1 Flash | 5368.1 | #35 |
| 2 | GPT-6.1 Sol | 3435.5 | #7 |
| 3 | GPT-6 Luna | 3019 | #52 |
| 4 | Claude Opus 5.5 | 2702.3 | #4 |
| 5 | GLM-5.3 Flash (High) | 2546 | #63 (GLM-5.3 Flash) |
| 6 | MiMo-V2.6-Flash | 2421.2 | #65 |
No result here: #6 GPT-6 Astra, #8 Gemini 4 Argon, #9 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=holdembench · How It Works · Data refreshed daily, snapshot 2026-10-18.