0:26
Before you trust any benchmark headline, ask one question first: who scored it? Today's board had the answer built right in: Nvidia ran Claude Opus 5 inside its own harness and posted a jump from 30% to 100% on the ARC-AGI-3 benchmark, scored by the same team that built the harness (via TechCrunch).
The read our prep pulled up, and I said it on air: "They just gave themselves an A on their own test." My co-host Nico pushed back immediately: ARC-AGI is a real third-party benchmark, not something Nvidia invented, so the real issue isn't the test, it's who's holding the pencil.
Where this came from
Value-First AI Daily - Aug 24, 2026Watch the full conversation this clip was pulled from.
The next step