By declaring that GPT-6 Astra has ushered in the AGI era, OpenAI is being flippant and cementing the term's status as nothing more than marketing
M.G. Siegler /Spyglass:NEW
Context & Ripple Effects
OpenAI’s launch paired an “AGI era” declaration with measurable product claims: computer-use tests involving DMV appointments and job searches, broad availability across its products, and a training run using more than 100,000 GPUs. Siegler’s critique targets the gap between those concrete claims and an expansive label.
The launch also puts OpenAI in a direct commercial comparison with Anthropic: GPT-6 Astra is priced to match Claude Fable 5.1, making capability evidence and reliability more actionable for buyers than an AGI designation.
First-order effects
- OpenAI’s AGI framing becomes a point of contention rather than a settled product attribute, concentrating scrutiny on Astra’s published capabilities, safeguards, and real-world computer-use performance.
Second-order effects
- Enterprise buyers comparing Astra with Claude Fable 5.1 gain little procurement guidance from the AGI label; matched pricing shifts the comparison toward task performance, controls, and deployment fit.
Third-order effects
- If major model vendors use “AGI” without a shared operational threshold, the term risks becoming launch positioning while benchmarks and safety evaluations become the practical basis for differentiation.
The trend: Frontier-model competition is turning grand intelligence claims into a marketing layer over a more concrete contest in agent performance, safeguards, and price.