For physicians
Get paid to be the hardest examiner an AI will ever face.
BenchEval is building a network of verified, specialty-matched physicians who evaluate frontier AI on real clinical reasoning. Remote, flexible, and demanding in the way good clinical work is.
Why join
Clinical judgment is the scarce resource in AI. Yours.
Work that uses your actual training
Not survey-filling. You'll write hard cases, grade model reasoning against your own, and hunt for the failure a layperson would never catch. The work is interesting precisely because it's difficult.
Shape what these systems become
Frontier models are learning medicine now, from whoever shows up to teach them. Physician judgment in the evaluation loop is how these tools become safe — and without it, they won't be.
Fits around clinical life
Remote, asynchronous, and task-based. Take work when you're available, decline when you're on nights. No minimum hours to stay in the network.
Founded by one of you
BenchEval is run by a practising emergency physician who still works shifts. The tasks, rates philosophy and review standards are set by someone who does this work too.
How it works
From application to first project
Apply
Tell us who you are: specialty, licence, experience, languages, availability. Ten minutes, no cover letter.
Verification
We verify your licence and credentials before any work reaches you — it's what makes 'verified physicians' true when we say it to buyers.
Calibration
A short paid exercise on sample tasks, so you know what the work feels like and we know how you grade before real projects start.
Specialty-matched work
Projects come to you matched to your specialty and interests — case writing, output grading, red-teaming — with clear briefs and clear rates before you accept.
Rates philosophy
Specialist work, priced like specialist work
We don't publish a rate card yet, and we won't invent one to look established. What we will commit to: rates are set for physicians, not for generic annotators — the comparison point is your clinical hour, not a data-entry wage.
Every task shows its rate before you accept it. No opaque bundles, and no unpaid "qualification" work — even the short calibration exercise is paid.
If a rate ever feels wrong for the work, we want to hear it. The person setting rates does these tasks too.
Join the network
We're building our physician network now.
Early applicants get first access to projects as they open — and a real say in how the network is built.
Apply as an Expert