MySI
Early prototype

Your own model benchmark

Find the AI that's best at your work.

Leaderboards test math olympiads. MySI tests the things you actually ask AI to do, runs them on three top models from three different companies, and lets you pick the best answers blind. You leave with one winner.

Copies one message. Paste it into your chat app and it sets MySI up for you, or walks you through it in about a minute.

ClaudeChatGPTGemini
Example blind test Texting
You

Text my landlord that rent will be two days late. Short, and don't grovel.

Your pick
AGPT-6.1 Sol
Hi, I wanted to reach out regarding this month's rent. Unfortunately, due to a delay with my paycheck, I will be approximately two days late. I sincerely apologize for any inconvenience this may cause.
Your pick
BClaude Sonnet 5.5
Hi, rent will be two days late this month. It'll be in by Friday. Thanks for understanding.
Your pick
CGemini 4 Argon
Hey! Just a heads up 🙂 Rent is going to be a couple days late, so sorry!! Should be there by Friday at the latest!
You picked B. It was Claude Sonnet 5.5.

Three steps, all inside your chat app

No new app to learn. MySI plugs into the chat you already use and does the heavy lifting there.

  1. Step 1 · about 5 minutes

    Tell it what you use AI for

    Your chat app asks a few questions and turns your real work into test tasks, each with what a great answer looks like to you.

  2. Step 2 · runs on its own

    Three models answer, blind

    GPT-6.1 Sol, Gemini 4 Argon and Claude Sonnet 5.5 each answer every task. You see them only as A, B and C, so the brand can't sway you.

  3. Step 3 · a click per task

    You pick, it names a winner

    Pick the best and worst answer for each task. Then the names are revealed and MySI tells you which model to use.

Why not just check a leaderboard?

A new model ships every week, and each one claims the top spot on someone's chart. None of those charts were made from your work.

A public leaderboard

  • Scores competition math and coding puzzles
  • Graded by whoever made the test
  • One ranking for everyone on earth

Your MySI benchmark

  • Tests the text to your landlord, the lesson plan, the spreadsheet formula
  • Graded by you, blind, against what you said a great answer needs
  • A ranking for one person: you

What you leave with

One winner you can act on today, even if you never switch apps.

Example result
Claude Sonnet 5.5

You picked it as best on 3 of 4 tasks, and it won both of your writing tasks.

In ChatGPT, your best option is GPT-6.1 Sol. Pick it there, then paste this into your custom instructions so it answers the way you rated:

Keep replies short and plain. No em-dashes. Lead with the answer, then one line of reasoning.
Your best picks
Claude Sonnet 5.5
3
GPT-6.1 Sol
1
Gemini 4 Argon
0

Later, ask your chat app "which AI should I use for this?" and MySI answers from your own results.

Adding it to your app

The message you copy carries these steps, so your app can walk you through them. Here they are too.

Claude: In claude.ai or the Claude app: Settings, then Connectors, then Add custom connector. Name it MySI and paste your link.

Then start a new chat and say Find my best AI model In Gemini: @MySI find my best AI model