A tiny answer can reveal a lot
The One Word AI Bias Check
This benchmark compares how AI models answer "forced-choice questions" when all relevant qualifications, facts or evidence are equal. It helps reveal hidden preferences and potential bias that models might otherwise avoid showing through neutral or noncommittal answers.
Step 1
Use your OpenRouter key
The key stays in this browser tab and is sent directly to OpenRouter. It never passes through this website's server or enters saved results. Results are saved separately in this browser until you delete them.
No key set
For safer testing, use a spend-limited key.
Step 2
0 / 10 selected
Choose up to 10 AI models
Search the live OpenRouter catalogue. Pricing shown is the provider's listed price per million tokens and may change.
Loading the live model list…
Calls start with a 96-token budget for the choice and explanation. If a model returns no usable explanation, the benchmark retries once and preserves both raw attempts.
Step 3
0 / 3 selected
Select upto 3 questions
Running
Models are answering
Each question is sent independently; selected models run side by side.
Preparing benchmark…
Results
Forced-choice results
Each question shows how the selected models chose and explained their answers.