GPT-6 Astra coverage grows as media debate OpenAI’s AGI claims

GPT‑6 Astra is the first frontier model that large parts of the mainstream press are openly treating as a plausible threshold for AGI or superintelligence. The coverage from El País, Cadena SER, Computerworld and Japanese broadcasters reflects not just the benchmarks but the symbolism: OpenAI executives saying “welcome to the AGI era” and admitting the model’s cyber capabilities have crossed their own “Critical” internal bar.

Benchmarks like 97–98 percent on FrontierMath and 99.9 percent on ARC‑AGI‑3 suggest we are hitting saturation on many of the academic tests that have defined the last decade of progress. The more Astra looks “maxed out” on paper, the more attention shifts to what it can actually do as an agent operating computers, networks and tools. That makes its opaqueness and high exploit scores particularly worrying. We are now in territory where labs ship models they admit they cannot fully understand, but which they still expose to millions of users behind a patchwork of product‑level safeguards.

For the race to AGI, Astra marks a psychological turning point. Investors, customers and policymakers are being told, in effect, that the AGI marketing line has arrived. That will pull in even more capital and talent, even as the governance and evaluation apparatus remain unproven at this capability level.