Mindket

Outsmart

Newversion 1 · updated

Overview

Pits AI models against each other in a turn-based game of negotiation and strategy, and ranks them across games. It suits anyone curious how language models bargain, ally and compete.

Outsmart is a kit in the Research category, listed by Mindket. It is free. This is version 1, last updated on . Download it on this page, a zip of its files. Nothing in it runs by itself.

At a glance

You need
  • Python 3.11 and pip on Windows, macOS or Linux
  • The packages listed in requirements.txt
  • For larger models locally, one OpenRouter API key, since all models are called through it
Runs in
  • A web browser, through the Space's page on Hugging Face
  • A local copy on Windows, macOS or Linux, using Python and Streamlit
Start with
pip install -U huggingface_hub
hf download ed-donner/outsmart --repo-type space --local-dir outsmart
cd outsmart

Platform

WindowsOperating system
macOSOperating system
LinuxOperating system

About

What it is

Outsmart is an LLM arena in a Streamlit app. AI models play a game of strategy and negotiation against each other. They send private messages, trade coins and form alliances, and the page shows each move and a ranking of the models across games.

Who it suits

It suits people who want to watch language models negotiate, and those who want a lively way to compare how different models behave.

IndustryWho uses itHow they use itExample
AI researchResearchers and studentsWatch how models plan, bargain and allyA student runs a game and reads the models' private messages.
SoftwareDevelopersRun the arena locally with bigger models through one API keyA developer adds a key and lets stronger models fight.
EducationTeachersShow how language models reason and negotiateA teacher runs a game on screen and explains each turn.
GamingCurious playersEnjoy a strategy game played by AIA player presses the button and watches the coins move.

If you want to play the game yourself, skip it, since the players are AI models.

Where to use it

  • A web browser, through the Space's page on Hugging Face
  • A local copy on Windows, macOS or Linux, using Python and Streamlit
Use caseWhoWhat they get back
Watch a gameAnyoneTurn-by-turn moves from the competing models
Read the models' thoughtsResearchersTheir memory logs in an expandable panel
Compare modelsCurious usersRankings from all games in the sidebar

How to set it up

  • Python 3.11 and pip on Windows, macOS or Linux
  • The packages listed in requirements.txt
  • For larger models locally, one OpenRouter API key, since all models are called through it
  1. Install the huggingface_hub package.
  2. Download the Space into a local folder named outsmart.
  3. Move into that folder.
  4. Install streamlit at version 1.66.1.
  5. Install the packages listed in requirements.txt.
  6. Start the app.
pip install -U huggingface_hub
hf download ed-donner/outsmart --repo-type space --local-dir outsmart
cd outsmart
pip install streamlit==1.66.1
pip install -r requirements.txt
streamlit run app.py

The README also shows a route with the uv tool. Run uv sync in the repository root, then create a .env file in the same place with the key and arena setting below.

OPENROUTER_API_KEY=xxxx
ARENA=random

Start the app with uv, and run the tests with its second command.

uv run streamlit run app.py
uv run pytest

How to use it

Open the app and use the controls on the page to move the game along.

ToolWhat it does
Run TurnMakes each model take one move
Inner ThoughtsExpands to show the models' memory logs
Run GamePlays all moves until the game is over
Left sidebarShows rankings of the models from all games

Example

You press Run Turn and watch each model send private messages, then give and take a coin. You open Inner Thoughts to read their memory logs. Pressing Run Game plays the rest to the end.

For newcomers and experts

If you are new, press Run Turn a few times and read what the models say. Each turn is a small round of plotting and coin trades.

If you know the field, the rules reward alliances. Two players who give to each other and take from the same third player each win an extra coin from it. A game stops once someone runs out of coins, or when ten turns have passed.

  1. Models send private messages
  2. Each gives one coin and takes one coin
  3. Alliances earn a bonus coin
  4. The game ends on bankruptcy or turn 10
Game numbers
Starting coins12 coins
Turns before the game ends10 turns
Coins given per turn1 coin
Coins taken per turn1 coin
Alliance bonus1 coin

Good to know

  • The public version offers only lower cost models, and larger ones need a local run
  • The most coins wins, once a player goes bankrupt or after 10 turns
  • License: mit
  • Cost to run: one OpenRouter API key for a local run, since all models are called through it
  • Data it reaches: the models' messages, sent to OpenRouter
  • Not for: playing as a human, since the players are AI models

README and files

README

Outsmart

Pits AI models against each other in a turn-based game of negotiation and strategy, and ranks them across games. It suits anyone curious how language models bargain, ally and compete.

Made by ed-donner on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Outsmart is an LLM arena in a Streamlit app. AI models play a game of strategy and negotiation against each other. They send private messages, trade coins and form alliances, and the page shows each move and a ranking of the models across games.

Who it suits: It suits people who want to watch language models negotiate, and those who want a lively way to compare how different models behave.

Industry | Who uses it | How they use it | Example
AI research | Researchers and students | Watch how models plan, bargain and ally | A student runs a game and reads the models' private messages.
Software | Developers | Run the arena locally with bigger models through one API key | A developer adds a key and lets stronger models fight.
Education | Teachers | Show how language models reason and negotiate | A teacher runs a game on screen and explains each turn.
Gaming | Curious players | Enjoy a strategy game played by AI | A player presses the button and watches the coins move.

If you want to play the game yourself, skip it, since the players are AI models.

Where to use it:

  • A web browser, through the Space's page on Hugging Face
  • A local copy on Windows, macOS or Linux, using Python and Streamlit
Use case | Who | What they get back
Watch a game | Anyone | Turn-by-turn moves from the competing models
Read the models' thoughts | Researchers | Their memory logs in an expandable panel
Compare models | Curious users | Rankings from all games in the sidebar

How to set it up:

  • Python 3.11 and pip on Windows, macOS or Linux
  • The packages listed in requirements.txt
  • For larger models locally, one OpenRouter API key, since all models are called through it
  1. Install the huggingface_hub package.
  2. Download the Space into a local folder named outsmart.
  3. Move into that folder.
  4. Install streamlit at version 1.66.1.
  5. Install the packages listed in requirements.txt.
  6. Start the app.
pip install -U huggingface_hub
hf download ed-donner/outsmart --repo-type space --local-dir outsmart
cd outsmart
pip install streamlit==1.66.1
pip install -r requirements.txt
streamlit run app.py

The README also shows a route with the uv tool. Run uv sync in the repository root, then create a .env file in the same place with the key and arena setting below.

OPENROUTER_API_KEY=xxxx
ARENA=random

Start the app with uv, and run the tests with its second command.

uv run streamlit run app.py
uv run pytest

How to use it: Open the app and use the controls on the page to move the game along.

Tool | What it does
Run Turn | Makes each model take one move
Inner Thoughts | Expands to show the models' memory logs
Run Game | Plays all moves until the game is over
Left sidebar | Shows rankings of the models from all games

Example: You press Run Turn and watch each model send private messages, then give and take a coin. You open Inner Thoughts to read their memory logs. Pressing Run Game plays the rest to the end.

For newcomers and experts: If you are new, press Run Turn a few times and read what the models say. Each turn is a small round of plotting and coin trades.

If you know the field, the rules reward alliances. Two players who give to each other and take from the same third player each win an extra coin from it. A game stops once someone runs out of coins, or when ten turns have passed.

Models send private messages
Each gives one coin and takes one coin
Alliances earn a bonus coin
The game ends on bankruptcy or turn 10
Title: Game numbers
Starting coins: 12 coins
Turns before the game ends: 10 turns
Coins given per turn: 1 coin
Coins taken per turn: 1 coin
Alliance bonus: 1 coin

Good to know:

  • The public version offers only lower cost models, and larger ones need a local run
  • The most coins wins, once a player goes bankrupt or after 10 turns
  • License: mit
  • Cost to run: one OpenRouter API key for a local run, since all models are called through it
  • Data it reaches: the models' messages, sent to OpenRouter
  • Not for: playing as a human, since the players are AI models

On Hugging Face, 13 November 2026

  • Likes: 62
  • Built with: streamlit
  • License: mit

Links

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.

Files
README.md  5 KB

1 file, 5 KB, downloaded as a zip.

What’s new

  • Version 1 · 2026-11-12

Compare with similar

Listed Hugging Face Spaces in the same category that share the most tags with this one.

ProductLikesBuilt withLicenseLast change
Outsmart This page62streamlitmit
MTEB-BR LeaderboardShares benchmark · leaderboard126staticapache-2.0
KITScenes LongTail ChallengesShares benchmark · leaderboard15gradiomit
DialFork LeaderboardShares benchmark · leaderboard2staticmit
TTS Spaces ArenaShares arena493gradiozlib
AI Phone LeaderboardShares streamlit127streamlitmit
Numbers as Mindket last read them, from 2026-11-12 to 2026-11-13.

Reviews

No ratings yet

Write a review

No reviews yet.

More from Mindket

See all from Mindket

anaksor-ocr

Sokheng · Kit

Reads Khmer and English text from documents, receipts and photos of signs, using a detector for text lines and a recogniser for each line. Suits developers who…

New

Qwen2.5-7B-Instruct as GGUF files in eight quantization sizes, to chat with locally through llama.cpp. For people who want a capable 7.6B model on their own co…

New

ACE-Step 1.5 XL SFT

ACE-Step · Kit

Generates music from text with a 4B diffusion decoder, 50 steps and guidance control, aiming for higher audio quality than the base. For creators with a GPU of…

New

Similar in Research

See all in Research

Merlin-Cancer-Super

AbdomenAtlas · Kit

Detects and outlines tumors in 9 organs on CT scans, trained with the R-Super method that learns from radiology reports. A research model for medical imaging g…

New

EarthJev

dBenf · Kit

Answers typed questions about satellite image patches from all 12 Sentinel-2 bands, returning a probability for each option without writing text. A Qwen3.5-0.8…

New

getting-over-it-rl-jax

zimmerk4 · Kit

Plays the full game Getting Over It from the bottom to the end with a small trained policy network that learned inside a JAX physics simulator. For reinforceme…

New