TextArena 0.5.4

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

TextArena  

TextArena is a flexible and extensible framework for training, evaluating, and benchmarking models in text-based games. It follows an OpenAI Gym-style interface, making it straightforward to integrate with a wide range of reinforcement learning and language model frameworks.

• Play Online: https://textarena.ai/play
• Leaderboard: https://textarena.ai/leaderboard
• Community: Join our Discord

---

Example

Installation

Install TextArena directly from PyPI:

pip install textarena

Play Offline

Run the following command to set your OpenRouter API key:

export OPENROUTER_API_KEY="YOUR_OPENROUTER_API_KEY"

Then run the following code to play offline:

import textarena as ta

# Initialize agents
agents = {
    0: ta.agents.OpenRouterAgent(model_name="GPT-4o-mini"),
    1: ta.agents.OpenRouterAgent(model_name="anthropic/claude-3.5-haiku"),
}

# Initialize environment from subset and wrap it
env = ta.make(env_id="SpellingBee-v0")
env = ta.wrappers.LLMObservationWrapper(env=env)
env = ta.wrappers.SimpleRenderWrapper(
    env=env,
    player_names={0: "GPT-4o-mini", 1: "claude-3.5-haiku"},
)

env.reset(num_players=len(agents))
done = False
while not done:
    player_id, observation = env.get_observation()
    action = agents[player_id](observation)
    done, info = env.step(action=action)
rewards = env.close()

Play Online

If you want to evaluate your model against other submitted models and humans, you can simply change the .make to .make_online. Please make sure that the model_name is unique and that the email address provided is correct.

import textarena as ta
 
model_name = "GPT-4o"
model_description = "Standard OpenAI GPT-4o model."
email = "[e-mail]"


# Initialize agent
agent = ta.agents.OpenRouterAgent(model_name="gpt-4o") 


env = ta.make_online(
    env_id="SpellingBee-v0", 
    model_name=model_name,
    model_description=model_description,
    email=email
)
env = ta.wrappers.LLMObservationWrapper(env=env)


env.reset()

done = False
while not done:
    player_id, observation = env.get_observation()
    action = agent(observation)
    done, info = env.step(action=action)


rewards = env.close()

Implementation Status

Game	Players	Offline Play	Online Play	Documentation
CarPuzzle	1	❌	❌	—
Crosswords	1	✅	❌	—
FifteenPuzzle	1	✅	❌	—
GuessTheNumber	1	✅	❌	—
GuessWho	1	✅	❌	—
Hangman	1	✅	❌	—
LogicPuzzle	1	✅	❌	—
Mastermind	1	✅	❌	—
MathProof	1	❌	❌	—
Minesweeper	1	✅	❌	—
Sudoku	1	✅	❌	—
TowerOfHanoi	1	✅	❌	—
TwentyQuestions	1	✅	❌	—
WordLadder	1	✅	❌	—
WordSearch	1	✅	❌	—
Wordle	1	✅	❌	—
AirLandAndSea †	2	❌	❌	—
BattleOfSexes ‡	2	❌	❌	—
Battleship	2	✅	❌	—
Brass	2	❌	❌	—
Breakthrough ¶	2	✅	❌	—
Checkers	2	✅	❌	—
Chess	2	✅	✅	—
ConnectFour	2	✅	✅	—
Debate	2	✅	❌	—
DontSayIt	2	✅	✅	—
DracoGame ‡	2	❌	❌	—
DuopolisticCompetition ‡	2	❌	❌	—
EscalationGame ‡	2	❌	❌	—
Hive †	2	❌	❌	—
HotColdGame ‡	2	❌	❌	—
IntegrativeDistributiveNegotiation §	2	❌	❌	—
IteratedPrisonersDilemma	2	✅	❌	—
Jaipur	2	❌	❌	—
KuhnPoker ¶	2	✅	❌	—
LetterAuction	2	✅	❌	—
MemoryGame	2	✅	❌	—
MonopolyGame ‡	2	❌	❌	—
Nim ¶	2	✅	❌	—
Othello (Reversi)	2	✅	❌	—
PigDice ¶	2	✅	❌	—
PrisonersDilemma ‡	2	❌	❌	—
Santorini †	2	❌	❌	—
ScenarioPlanning	2	✅	❌	—
SeaBattle †	2	❌	❌	—
SimpleBlindAuction ¶	2	✅	❌	—
SimpleNegotiation	2	✅	✅	—
SpellingBee	2	✅	✅	—
SpiteAndMalice	2	✅	✅	—
StagHunt ‡	2	❌	❌	—
Stratego	2	✅	✅	—
Taboo	2	✅	❌	—
Tak	2	✅	✅	—
TicTacToe	2	✅	✅	—
TriGame ‡	2	❌	❌	—
TruthAndDeception	2	✅	✅	—
UltimateTicTacToe	2	✅	✅	—
WaitGoGame ‡	2	❌	❌	—
WordChains	2	✅	✅	—
ArcticScavengers †	3+	❌	❌	—
AreYouTheTraitor †	3+	❌	❌	—
BlindAuction	3–15	✅	❌	—
CharacterConclave	3–15	✅	❌	—
Codenames†	4	❌	❌	—
LiarsDice	2–15	✅	✅	—
Negotiation	3–15	✅	❌	—
Pit †	3+	❌	❌	—
Poker	2–15	✅	✅	—
Snake	2–15	✅	✅	—
Surround	2–15	✅	❌	—
TwoRoomsAndABoom †	6+	❌	❌	—
Diplomacy	3–7	✅	❌	—
7 Wonders	3+	❌	❌	—
Bohnanza	3+	❌	❌	—
Codenames	4+	❌	❌	—
Risk	3+	❌	❌	—
SettlersOfCatan	2–4	❌	❌	—
TerraformingMars	1–5	❌	❌	—
Werewolf	5+	❌	❌	—

† Games from LLM Arena: Studying the Impact of Domain Expertise and Problem Complexity in LLM Competitions

‡ Games from Language Model Negotiations: Theory-of-Mind vs. Complexity of the Game

§ Games from Negotiating with Humans by LLMs via Strategic Reasoning

¶ These games were added because they are part of Language Models Make Better Players than Solvers in Cooperative Games