TextArena/TextArena

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

What it solves

TextArena provides a standardized way to benchmark and train Large Language Models (LLMs) using text-based games. It addresses the need for environments where models can be tested on their ability to reason, strategize, and interact in multi-player scenarios without relying on traditional visual interfaces.

How it works

The framework uses an OpenAI Gym-style interface, allowing it to be integrated with various reinforcement learning (RL) and language model frameworks. It provides a suite of over 100 games (including single, two-player, and multi-player games) where agents—which only need a call function that takes a string observation and returns a string action—can compete or solve puzzles.

Who it’s for

It is designed for AI researchers and developers focusing on LLM training, evaluation, and the reasoning capabilities of models, particularly those interested in multi-agent reinforcement learning and theory of mind.

Highlights

  • Over 100 text-based games for benchmarking and training.
  • OpenAI Gym-style interface for easy integration with RL frameworks.
  • Support for single, two-player, and multi-player game environments.
  • Compatible with various LLM agents via a simple string-in, string-out interface.

Related

  • 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
    google-deepmind/open_spiel

    A framework for reinforcement learning and search/planning in games, providing a collection of environments and algorithms for multi-agent and single-agent research.

  • 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.

  • Project
    TsinghuaC3I/MARTI

    An open-source framework for training LLM-based multi-agent systems using reinforcement learning, featuring tree search augmentation for complex reasoning and code generation.

  • Project
    OpenManus/OpenManus-RL

    An open-source framework for optimizing LLM agents using reinforcement learning to enhance their reasoning, tool use, and decision-making capabilities.