Theo Spent $330,000 Testing GPT 6 — Here’s What We Think

Theo says he’s put roughly $330,000 worth of inference through GPT-6 Astra — so we watched his review, reacted to the claims, and then ran some tests of our own. We get into GPT-6 vs Claude/Fable, token pricing, coding quality, benchmark gaming, ARC-AGI, reasoning effort, and whether GPT-6 actually feels like a generational leap. Then we put GPT-6 Astra and Fable head-to-head on the same coding challenge: build the wildest version of Pong possible. Which model actually did better? And is GPT-6 really as “next-gen” as the hype suggests? Subscribe for more AI reactions, model comparisons, experiments, and practical testing.