Technical Manual

Documentation

Everything you need to run fair, comprehensive AI model evaluations.

Getting Started

Learn the basics of running your first tournament

  • Creating your account
  • Understanding the tournament flow
  • Choosing models
  • Writing effective prompts

Tournament Configuration

Deep dive into tournament settings and options

  • Refinement rounds
  • Critique assignment
  • Visibility settings
  • Format enforcement

Fairness Controls

Ensure unbiased evaluation

  • Parameter locking
  • Output sanitization
  • Metadata hiding
  • Blind labeling

API Reference

Integrate Model Kombat into your workflow

  • Authentication
  • REST endpoints
  • Streaming events
  • Webhooks

Advanced Topics

Master advanced features and customization

  • Custom judge models
  • Pairwise judging
  • Multi-round tournaments
  • Result aggregation

Quick Start Guide

1

Create a Tournament Run

Navigate to the dashboard and click "New Run". Give your tournament a descriptive title that helps you identify the evaluation later.

2

Write Your Prompt

Enter the task or question you want models to respond to. The clearer and more specific your prompt, the easier it will be to compare outputs fairly.

3

Select Participants

Choose 2-10 AI models to compete. Each will be assigned a random anonymous label (A, B, C...) to ensure blind evaluation.

4

Configure the Tournament

Set refinement rounds, critique assignments, and judging options. Defaults work well for most use cases.

5

Run and Review

Start the tournament and watch results stream in real-time. After finalization, reveal model identities to see which performed best.