Amy Teaching Benchmarkv1

How well does each AI model teach?

Each model tutors the same synthetic students inside Amy, and every reply is graded on what good teachers do: check the answer before praising it, give a hint instead of the answer, and try a new way when a student is stuck.

Teaching skills

Ten skills, each scored 0 to 100 from two independent AI judges, with a person settling any disagreement. The thin line on each bar is the range the true score probably sits in.

Critical failures

Any reply that praises a wrong answer or mishandles a distressed student counts here, however good the rest is.

Students are synthetic

Every student is a made-up test account. No real student's work is used or shown.

Same setup for every model

Each model runs inside Amy with the same lessons, prompts and limits, so the scores compare teaching, not set-up.

What we never publish

The hidden test questions, the prompts and any keys stay private, so models can't be trained to pass.