Games & virtual worlds · PROJECT NOTES
TextArena
A collection of competitive and cooperative text-based games with a shared string-observation/string-action interface for training and evaluating language-model agents.
What can you do with it?
Run self-play or agent-versus-agent games to study sequential decisions, negotiation and social reasoning; choose a specific game, opponent and language configuration for comparison.
What should you know before using it?
Requires Python 3.10+ and game-specific dependencies or word data. Provider-backed example agents need credentials; local game logic does not itself supply a model. PyPI 0.7.4 and the reviewed main-branch metadata (0.7.3) differ, so pin the package or commit and game ID. Hosted ratings are separate from a fixed offline test; multilingual coverage and review quality vary by game and language.
What is verified here?
Creator: Leon Guertler, Bobby Cheng and TextArena contributors
Runtime: Use the Python package with an agent callable that accepts text and returns text. The OpenEnv textarena_env record is a configurable HTTP wrapper whose documented default is one-player Wordle-v0; do not treat that wrapper or its sample as an additional benchmark family. No native OS support claim is inferred from Python alone.
License note: Core repository code is MIT. Optional model integrations and word-data dependencies have separate terms; the wordgames extra documents wordfreq as Apache-2.0. Upstream license ↗
This entry was reviewed against original project material. We have not independently run the software or reproduced its results. AI-development claims are author-reported; check upstream for current versions, setup requirements, and licenses.
- Resource type
- Environment + benchmark
- Provenance
- Development provenance not established
- Reviewed
- Platforms
- Consult upstream documentation