PokerSkill: LLMs Can Play Expert-Level Poker without Training or Solvers

2026-05-28 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Gaming & Interactive Media · Depth: Expert, quick

Summary

The PokerSkill framework enables Large Language Models (LLMs) to play expert-level poker without requiring extensive training or reliance on equilibrium solvers. Developed as a training-free and solver-free solution, PokerSkill addresses the limitations of both traditional rule-based agents and direct LLM play, which typically fall far below optimal performance. It integrates LLMs with a detailed, human-expert-designed rule-based skill library, using a deterministic context engine to analyze game state and retrieve relevant skill fragments, thereby constraining the LLM's action choices to reasonable options. Against GTOWizard, a state-of-the-art GTO benchmark, GPT-5.5 XHigh with PokerSkill achieved -57 ± 21 mbb/hand, Claude Opus 4.6 achieved -80 ± 29 mbb/hand, and Claude Opus 4.7 achieved -87 ± 64 mbb/hand. This performance reduced losses by 49-61% compared to default-prompt baselines and surpassed the strong bot Slumbot, marking the first instance of an LLM achieving competitive play in a complex imperfect-information game without game-specific training or solver queries.

Key takeaway

For Machine Learning Engineers developing AI for complex imperfect-information games, PokerSkill demonstrates a viable path to expert-level performance without the immense computational cost of traditional solver-based training. You should consider integrating expert-designed rule systems as structured action-grounding interfaces for LLMs. This approach allows you to deploy competitive agents in domains like strategic decision-making or resource management, significantly reducing development time and computational resources compared to training large game-specific models.

Key insights

Combining LLMs with structured, rule-based poker skills enables expert-level play without training or solvers.

Principles

• Rule-based skills alone do not constitute a strong strategy.
• LLMs alone cannot play complex imperfect-information games well.
• Hybridizing LLMs with expert rules yields competitive, training-free agents.

Method

PokerSkill employs a deterministic context engine to analyze game state, retrieving relevant fragments from a layered, human-expert-designed skill library to constrain LLM action choices.

In practice

• Integrate expert-designed rule systems to ground LLM actions.
• Develop training-free agents for complex games using LLM-rule hybrids.
• Explore structured interfaces for LLM decision-making in imperfect-information settings.

Topics

• PokerSkill
• Large Language Models
• Imperfect Information Games
• Training-Free AI
• Rule-Based Systems
• Game AI

Code references

• lbn187/PokerSkill

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• Building a Chess Coach — Anant Dole and Asbjorn Steinskog, Take Take Take — Combining classical chess engines with LLMs and human-like move prediction enhances AI chess coaching.
• OptSkills: Learning Generalizable Optimization Skills from Problem Archetypes via Cluster-Based Distillation — OptSkills learns generalizable optimization skills by clustering problem archetypes and distilling successful workflow trajectories.
• Training-Free Agentic AI: Probabilistic Control and Coordination in Multi-Agent LLM Systems — Probabilistic control with belief-guided delegation significantly boosts multi-agent LLM system efficiency and robustness without training.
• PTCG-Bench: Can LLM Agents Master Pokémon Trading Card Game? — LLM agents show promise in complex games but struggle with sustained self-evolution and are sensitive to evaluation harness design.
• [P] Vibecoded on a home PC: building a ~2700 Elo browser-playable neural chess engine with a Karpathy-inspired AI-assisted research loop — An AI-assisted research loop can rapidly develop high-performance neural chess engines on consumer hardware.
• AI-Enabled Serious Games: Integrating Intelligence and Adaptivity in Training Systems — AI can integrate intelligence and adaptivity into serious games, addressing static design and limited learner modeling.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.