Decisions

DecisionsA Launch-Day AI Benchmark Is Not a Fixed Number

A Launch-Day AI Benchmark Is Not a Fixed Number

OpenAI revised several of GPT-6 Astra's published evaluation scores, and a rival's scores, in the hours after its September 3 launch. What the edits were, and why any procurement decision based on a launch-day benchmark card needs a dated, hashed snapshot.

3 min read

Page 1 / 11