MARK-17 Documentation
Everything you need to install MARK-17, connect the AI models you want to test, run a benchmark, and understand what the results mean.
New to MARK-17?
Start with the Quick Start Guide, then set up whichever kind of model you want to test first — local or cloud. The wizard walkthrough is worth reading before your first real benchmark.
Open the Quick Start Guide (PDF)Getting started
Install it, run your first benchmark, and understand what came back.
Quick Start Guide
Install MARK-17 and run your first benchmark from a standing start.
Open PDF →The Benchmark Wizard
A walkthrough of the wizard, from describing your task to reading live results.
Open PDF →Understanding Your Score
What the composite score means, how domain breakdowns work, and how to read a result.
Open PDF →Setting up models
Connect the AI you want to test — on your own hardware, in the cloud, or both.
Setting Up Local Providers
Configure Ollama, LM Studio, or the built-in runtime to benchmark models on your own hardware.
Open PDF →Cloud Provider Setup
Connect OpenAI, Anthropic, Google Gemini, Groq, and other cloud APIs using your own keys.
Open PDF →Cost Estimator
Understand API and electricity costs before you run, and produce a cost analysis.
Open PDF →Reference
The complete manual and the detail behind what gets tested.
User Manual
The complete reference guide, suitable for reading offline.
Open PDF →Test Suites Reference
Every built-in test suite — what it measures, which domains it covers, and when to use it.
Read →MBX Format Specification
How tamper-evident benchmark exports are structured and verified.
Read →Troubleshooting and help
Common questions
Local versus cloud, hardware requirements, where your data goes, what MBX and Advisor are, and what you can export.
Read the MARK-17 FAQ →Still stuck
Installation problems, unexpected behaviour, and licensing questions each have their own route so you reach the right person.
Get support →Documentation is actively being expanded. If a guide you need does not exist yet, tell us at support@aibenchlab.com and we will prioritise it.