Big brain.
Good boyfriend?

You understand attention mechanisms. But can you pay attention?

Find your boyfriend model tier in 12 questions, then collect your BF-Bot card.

12 questions~3 minutes
BOYFRIENDBENCH#003
Rare BF-Bot

Sonnet

72/100
Sonnet: a pink desktop computer robot with a heart antenna and a waving hand.
Good boyfriend. In preview.High intelligence. Occasionally needs a system prompt.
CONTEXT 78EMPATHY 67EFFORT 78REPAIR 67
Conditional thoughtfulness

Very thoughtful once the requirements are clarified.

Self-reported scoreFor fun ♥
EXAMPLE CARD

Love is in the little evals.

Four categories, equally weighted.

01

Context retention

The coffee order. The big meeting. The name of that one friend’s dog.

02

Emotional bandwidth

Sometimes “that sucks, babe” beats a 40-minute root cause analysis.

03

Autonomous effort

Planning the date. Folding the laundry. No elaborate system prompt.

04

Repair & alignment

Acknowledging the bug, owning the impact, and actually shipping a fix.

What are you running on?

Your score determines your model tier.

Swipe to meet all five BF-Bots →

Go on. Check your weights.

BOYFRIENDBENCH / HOW SCORING WORKS

Peer review pending.

Inspired by one devastating sentence: “That’s why you’re only scoring Sonnet on BoyfriendBench.”

The scoring function

Twelve scenarios, four equally weighted categories. Each answer earns 0–3 points. Your total out of 36 becomes a score out of 100; category scores use the same scale.

Privacy / legal

Privacy

Your answers and optional card name stay in this browser tab. Progress survives a refresh.

We count quiz starts, questions reached, completions, and card-sharing or download actions in daily totals. These counts contain no names, answers, scores, card images, or visitor identifiers.

Sites also provides basic traffic analytics. Restricted previews use ChatGPT sign-in for access. Your shared card includes only your chosen name and results; the caption includes a link to the quiz.

For fun

Model names are playful fictional tiers, not an assessment of real AI models. This isn’t relationship advice, a validated test, or an excuse to rank actual humans.