We have a new king of the hill: GPT-5. - 94.6% on competition math (AIME 2025) - Builds working websites/apps in ONE prompt - 80% fewer errors vs previous models - Outperforms experts in ~50% of professional tasks Unified system: Auto‑routes between quick answers and deep “thinking” when needed (or say “think hard” to force it). Reliability: ~45% fewer errors vs GPT‑4o; ~80% fewer vs o3 when reasoning. Less sycophancy and clearer about limits.

@SoloToCEO impressive benchmarks, but kinda sad how little this actually changes things

@SoloToCEO I'd like to know it's memory retention