# Blind model test

A skill based on a story on Agent Stories.

Sets up a small blind test of two AI models on your real task so you pick on quality and cost, not habit.

Task: Build with a coding agent
Works with: Claude, ChatGPT, Gemini
Tools: Claude Haiku 5.5, Gemini 3.8 Flash, Asana
Difficulty: Advanced
Country: United States

Use this when you want a small blind test of two AI models on your real task, so you pick on quality and cost.
Skip this if you only have one model to try, or you want a small test treated as proof.

Credit: From the story of Micah Berkley. An agency CTO blind-tested a cheaper AI model on 10 client calls and cut one AI bill by 74% while catching more promises
Source: https://agentstories.org/stories/micah-berkley-blind-model-test-cut-bill.html

## Skill text

Please turn the text below into a reusable skill. If you are Claude, save it as a skill. If you are ChatGPT or Gemini, save it as the instructions for a Project, custom GPT or Gem. If you cannot save skills, just follow these instructions now.

Name: Blind model test

Description: Sets up a small blind test of two AI models on your real task so you pick on quality and cost, not habit.

Ask me for:
1. The task.
2. 10 real examples with the right answers I wrote.
3. The two models and their prices.

Steps:
1. Write a scoring sheet: correct, missed, made up.
2. Label outputs A and B so I grade blind.
3. Total the scores and cost per task.

Output:
1. The scoring sheet.
2. A results table with score and cost per model.

Rules: Do not tell me which model wrote which output until I finish grading. Treat a small test as a hint, not proof.

## More

- [This skill as a web page](https://agentstories.org/skills/blind-model-test.html)
- [All skills, as a Markdown list](https://agentstories.org/skills.md)
- [Skills directory](https://agentstories.org/skills/)
