Give the models a task. Let them fight. Build with the winner.
LLM Fight Club is a terminal where AI agents fight against each other to find the best model for your specific task.
Instead of asking "Which LLM is the best?", you give the models the task and let them prove it.
Pick a task here. Choose your fighters. Start the fight. Watch them compete. Build with the winner.
Give the Fight Club any task you want to solve.
Build a Rust CSV parser that handles
quoted fields, escaped quotes and malformed rows.
Include tests.
Multiple LLM agents receive the same task and work on it independently.
🥊 LLM FIGHT CLUB
Claude GPT Gemini Grok
│ │ │ │
└───────┴───────┴────────┘
│
YOUR TASK
Each agent produces its own solution.
The Fight Club evaluates the results using objective checks whenever possible:
- Tests
- Compilation
- Correctness
- Output validation
- Requirements
- Edge cases
- Code quality
For subjective tasks, agents can also be evaluated by a blind judge.
The fighters are scored against each other.
ROUND 1
CLAUDE █████████
GPT ████████
GEMINI ███████
GROK ██████ ← ELIMINATED
ROUND 2
CLAUDE █████████
GPT ████████ ← ELIMINATED
FINAL
CLAUDE 🏆
The final fighter is the model that performed best for your task.
╭─────────────────────────────╮
│ 🏆 WINNER │
│ │
│ CLAUDE │
│ │
│ SCORE 8.7 / 10 │
│ LATENCY 2.4s │
│ COST $0.018 │
│ TESTS 42 / 42 │
│ │
│ → BUILD WITH CLAUDE │
╰─────────────────────────────╯
There is no single best LLM.
One model might be better at coding.
Another might be better at reasoning.
Another might produce better writing.
And the model that wins on a benchmark might not be the model that performs best on your actual problem.
LLM Fight Club lets you test them against each other on the work you actually care about.
Don't argue about which model is better. Make them fight.
Pretty much anything:
CODING
→ Build a REST API
→ Fix a broken codebase
→ Write a parser
→ Optimize an algorithm
REASONING
→ Solve a complex problem
→ Analyze a system
→ Find the best approach
RESEARCH
→ Investigate a topic
→ Compare competing ideas
→ Analyze a dataset
WRITING
→ Write an article
→ Create marketing copy
→ Develop a story
AGENTS
→ Complete a multi-step task
→ Use tools
→ Navigate a workflow
The important part is that the task comes first.
The winner is determined by how well a model handles that particular task.
Everything happens inside the Fight Club terminal.
╔══════════════════════════════════════════════════╗
║ LLM FIGHT CLUB ║
╠══════════════════════════════════════════════════╣
║ ║
║ TASK ║
║ Build a Rust CSV parser ║
║ ║
║ FIGHTERS ║
║ ● Claude ║
║ ● GPT ║
║ ● Gemini ║
║ ● Grok ║
║ ║
║ [ START FIGHT ] ║
║ ║
╠══════════════════════════════════════════════════╣
║ ║
║ CLAUDE VS GPT ║
║ ║
║ ████████████ █████████ ║
║ ║
║ ROUND 2 ║
║ ║
╚══════════════════════════════════════════════════╝
The terminal gives you a visual overview of the fight, the current round, model performance, scores, and the eventual winner.
The best model depends on what you're actually trying to do.
Don't rely on assumptions about which model should win.
If a solution can be tested objectively, test it.
When human-like judgment is required, hide the model identity to reduce bias.
The "best" model isn't always the most capable one.
A cheaper, faster model that solves your task just as well may be the better fighter.
🚧 Early prototype
LLM Fight Club is actively being developed.
Current focus:
- Multi-model execution
- AI agent orchestration
- Fight / elimination system
- Objective verification
- Blind judging
- Cost tracking
- Latency tracking
- Terminal UI
- Winner selection
- Winner handoff
- Benchmark history
The models don't get to tell you who's the best.
You make them fight.
MIT