Benchmarks

How well can a system act for you?

Agent benchmarks ask if a system can complete a task. Shelf Bench asks if it made the right call for you that’s measured by what you do next.

PI Index

Personal intelligence, measured

Opus 5.5

Anthropic
GPT-5.6 Luna

OpenAI
GPT-6.1 Sol

OpenAI
DeepSeek V4.1 Flash

DeepSeek
Gemini 3.8 Flash

Google

Join the leaderboard.

The PI Index

Each evaluation comes from real people using real community-built apps. Every run is graded on what the person did next.

Score = ƒ Model Harness Context

2. Create Coming soon Can it make something they want? Your Daily Podcast Your Week in Memes
3. Recall Coming soon Retrieve and connect facts about them and their activity. Weekly Life Recap This Week Last Year
4. Act Coming soon Can it get something done for them? Date Night Booker Subscription Renewal Manager
5. Represent Coming soon Can it speak accurately on their behalf? The Warm Intro Birthday Gift Concierge