Outsmart — AI skill for Claude Code
An LLM Arena that pits AI models against each other in a battle of diplomacy and deviousness.
How to install Outsmart
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open ed-donner/outsmart and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Outsmart does
An LLM Arena that pits AI models against each other in a battle of diplomacy and deviousness.
Alternatives in AI
- Deepreasoning — A high-performance LLM inference API and Chat UI that integrates DeepSeek R1's CoT reasoning traces with Anthr 5.4k ★
- Manifest — Real-time cost observability for OpenClaw agents — track tokens, costs, messages, and model usage 4.1k ★
- Vllm Mlx — High-performance OpenAI and Anthropic compatible LLM inference server for Apple Silicon 1.5k ★
README
title: Outsmart emoji: 🧠 colorFrom: yellow colorTo: red python_version: 3.11 sdk: streamlit sdk_version: 1.66.1 app_file: app.py pinned: false license: mit fullWidth: true header: mini tags: ["arena", "benchmark", "leaderboard"] thumbnail: https://edwarddonner.com/wp-content/uploads/2024/08/outsmart.jpg
Outsmart
A battle of diplomacy and deviousness between LLMs
Outsmart is an LLM Arena that pits AI models against each other in a game of strategy and negotiation.
[Play the game](https://edwarddonner.com/outsmart/) [Read the backstory](https://edwarddonner.com/2024/08/06/outsmart/) on my website [Clone the repo](https://github.com/ed-donner/outsmart) to use your API keys and fight with frontier models!
Rules of the Game
- Each player starts with 12 coins
- With each turn:
- Players send private messages to all other players to hatch plans
- Players select to give 1 coin to 1 player, and take 1 coin from another player
- If 2 players give each other coins, and both take from the same player, that is considered an "alliance" and they each receive an additional coin, taken from the player that they ganged up on
- The objective is to negotiate with the other players and make coins. The winner is the player with the most coins when one player goes bankrupt, or after 10 turns.
Using the Arena
- Press the Run Turn button to see the LLMs make a move
- Open the Inner Thoughts expander if you want to see their memory logs
- Press the Run Game button to run all moves until the game is over
- Expand the left hand sidebar to see LLM rankings from all games
- For the public version, only lower cost LLMs are available to compete, otherwise I might go bankrupt! Follow the instructions below to run locally and try larger models.
Installing locally
All models are called through [OpenRouter](https://openrouter.ai), so you only need one API key.
- If you don't have it already, install uv
- git clone
Related Skills
LLM Model Ranker 2026
Which AI model should you actually use? A cross-validated ranking of 39 frontier models across 7 independent l
Harness Arena
Blind, LMSYS-style arena for comparing AI coding agent harnesses (Claude Code, Codex CLI, OpenClaw, Hermes, On
Harness Arena Fe
Blind, LMSYS-style arena for comparing AI coding agent harnesses (Claude Code, Codex CLI, OpenClaw, Hermes, On
Agent Arena
Open-source evaluation framework for LLM agents. Run head-to-head A/B tests, score with LLM-as-judge rubrics,
Model Check
Validate Claude model usage against current public models
Duet
Two AI models build it; the harness checks the rules. fix/add/refactor/build/plan where tests are replayed aga
Related Agents
Security Lens
Application security lens for DevBox reviews. Threat-models a plan or diff across its trust boundaries (STRIDE
Poteto Fable
Third-family writing delegate. Exists so arena and architect can fan out across genuinely different models ins
Factory Reviewer A
Adversarial review panelist A. First seat on interrogate, architect and arena panels. Reviews a design or diff