Early hands-on: GPT-6.2 wins many benchmarks and is surprisingly cheap

ramos_casals · x · 2026-11-18

Spanish-language tech account AIDrugsYRobots shares first impressions of GPT-6.2 (Astra and Sol): initial testing suggests it leads on many benchmarks and is surprisingly cheap. The authors call it "crazy" and promise a detailed review after more testing. No concrete numbers yet.

Original post →

More from Models

Models channel →