Desk note: how to read AI model launch claims without the hype

Model launch weeks are noisy. Wire coverage often repeats vendor slides before independent evals land.

This desk note is a short checklist for readers: contamination risk, contamination of public benchmarks, tool-use vs pure knowledge, and the difference between demo latency and production SLA.

Treat vendor charts as claims. Prefer primary papers, model cards, and third-party harnesses when they exist.

Frontier Flux is independent and not affiliated with any lab.