SemiAnalysis says Claude's $200 plan delivers 5.6x the value of ChatGPT's ($11,726 vs $2,084 in API-equivalent usage), why Nick burned through his $500 ChatGPT plan in about 15 minutes, and why cost per task may be a better yardstick than token value. Reflection's new 500B open-weight model Beam and its rough benchmarks. Four AI labs refuse, under oath at a New York City hearing, to guarantee their agents will follow safeguards, and how reinforcement learning makes that hard. Plus: a scammer tries to hijack our giveaway, is AI the modern Tower of Babel, and viewer Q&A on models, Dots, and AI critics. Watch on YouTube: https://youtu.be/4fog90h4WS0 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: reach out to media@stackedpod.io.