OpenAI says its newly launched GPT-6 Astra model brings humanity to the threshold of AGI, but external evaluators argue that inflated lab metrics and opaque testing obscure its true reliability.