This week, we're inviting teams to put Sapien's evaluation system for AI work through a direct test using their own data, quality standards and expert judgments.
The Short Version
Beat Proof of Quality and win a Tesla. US robotics, world model and physical AI teams can enter by submitting their hardest dataset and defining what good looks like. Each organization can enter once, at sapien.io/challenge.
Five teams, 72 hours each, with evaluation targets agreed before testing. The organization behind the first verified failure chooses a Tesla Model 3, $25,000, or a $25,000 donation to an eligible charity. Applications close at 5 p.m. Pacific on Wednesday, 28 October.
Reviewing 100 million items would take each of ten experts about 40 years. Our new article explains how experts set the standard on a sample, and how AI judges qualify against hidden expert answers before taking production seats.
Beat Proof of Quality. Win a Tesla.
US robotics, world model and physical AI teams can enter the challenge by submitting their hardest dataset and defining what good looks like. Each organization can enter once.
We'll select five teams and agree on evaluation targets before testing. Each challenge will run for 72 consecutive hours, with the five tests conducted in a fixed order.
The organization behind the first verified failure chooses one prize: a Tesla Model 3, $25,000, or a $25,000 donation to an eligible charity. If Sapien passes all five challenges, the prize goes unawarded.
Applications close at 5 p.m. Pacific on Wednesday, 28 October, or earlier if the prize is awarded.
If you know an eligible team, ask it to include your name and email on its application. The winning team's verified referrer receives $1,000 within three days of verification of the winner and referral. Refer a team →
Send us your hardest dataset and tell us what good looks like.
Ten experts, a hundred million data points
At 30 seconds per check, 100 million items require roughly 833,000 hours of review. Across ten experts, that's about 83,000 hours each, or 40 years of full-time work.
Our new article explains how experts establish the standard on a sample, AI judges qualify against hidden expert answers, and the strongest receive production seats.
Building With Sapien?
The community keeps shipping interesting work. If you are building an agent with a need for verified data, share it with us.
Apply to the Builder Circle and help shape what comes next for Proof of Quality.
Join our Discord to be part of updates, AMAs, and more!
Participation in the Sapien challenge is free and requires no purchase.
Tesla, Inc. is not a sponsor of, and is not affiliated with, this challenge. Model 3 is a trademark of Tesla, Inc.
