Every's Vibe Check: GPT-6.2 Astra and Sol handle bounded tasks but fall short on judgment

danshipper · x · 2026-11-19

Every published its Vibe Check of OpenAI's GPT-6.2 Astra and Sol, with a live re-test scheduled Nov 18 at 3pm ET. The verdict: both models handle bounded tasks efficiently, but the deep work requiring judgment and follow-through still demands more. The six-person testing team includes Dan Shipper, Katie Parrott, Jack Cheng, Kieran Klaassen and Mike Taylor; prior editions covered Gemini 4 Carbon, GPT-6 Astra and Fable 5.2.

Original post →

More from Models

Models channel →