The Model Scored 30%. The Harness Scored 100%. That Gap Is an Investment Thesis.
Nvidia's AVO framework lifted Claude Opus 5 from 30% to a perfect 100% on ARC-AGI-3 without touching the model. The finding isn't about a benchmark. It's empirical proof of where value accumulates in agentic AI — and what it means for application-layer builders in Brazilian fintech.
Read article