Article

Evaluating a Frontier Agent with EBench: From Direct Control to Harness with Policy

This report evaluates GPT-6-Astra as a robot agent on EBench, using the same visual and proprioceptive observation protocol as post-trained policies, and compares two configurations: direct control and harness with policy.

Part of the event: EBench evaluation of GPT-6-Astra controlling robots

This page provides a summary and source link only.