The candidate open models were not usable as the agent’s main reasoning loop, and were not safe as substitutes for the verification gates, when judged against Claude Opus.
What is covered in the Full Insight:
Executive Summary
Background
Problem 1: Holding The Multi-Step Workflow Together