TextArena/TextArena

A Collection of Competitive Text-Based Games for Language Model Evaluation and Reinforcement Learning

What it solves

TextArena provides a standardized framework for evaluating and training language-model agents. It addresses the lack of diverse, interactive environments where LLMs can be tested for reasoning, negotiation, and social deduction through competitive and cooperative play.

How it works

The framework uses an interface styled after OpenAI Gym, allowing users to plug in LLM agents that interact with over 100 text-based games. These games range from classic board and card games to complex multi-agent reasoning tasks. The system supports self-play and model-vs-model evaluations and can be presented in up to 192 different languages, allowing for cross-lingual agent evaluation.

Who it’s for

It is designed for AI researchers and developers who want to benchmark LLM agents, build training pipelines using reinforcement learning, or study theory of mind in language models.

Highlights

  • 100+ Games: Includes a wide variety of single-, two-, and multi-player environments.
  • Multilingual Support: UI localization for 192 languages, enabling agents to play in different languages independently.
  • Gym-like Interface: Easy integration for agents that only need to implement a __call__ function.
  • ** webinars/competitions**: Associated with the MindGames competition for training LLMs on tasks requiring theory of mind.

Related

  • Project
    microsoft/TextWorld

    A text-based game generator and sandbox environment used for training and testing reinforcement learning agents in simulated text worlds.

  • Dispatch
    TextQuests: Evaluating LLM Agentic Reasoning in Text-Based Video Games

    Hugging Face introduces TextQuests, a benchmark using 25 classic Infocom interactive fiction games to evaluate the long-context reasoning and exploratory capabilities of LLM agents.

  • Project
    langwatch/scenario

    An agent testing framework that uses simulations, user simulators, and judge agents to evaluate the behavior and robustness of AI agents in multi-turn conversations.

  • Project
    hkust-nlp/Toolathlon

    Toolathlon is a benchmark for assessing language agents' general tool-use capabilities across 600+ diverse tools in realistic, long-horizon software environments.

  • Project
    rllm-org/rllm

    An open-source framework for training language agents with reinforcement learning, allowing the same agent code to be used across different backends, sandboxes, and benchmarks.