GPT-4.5 = Big Model Energy | YC Decoded

TL;DR

GPT-4.5 is OpenAI’s largest model (~10× GPT‑4 size), excelling in emotional intelligence, natural conversation, creativity, and reduced hallucination (37% vs. GPT‑4o’s 61.2%). It improves on subjective tasks (writing, humor, irony) but underperforms compared to reasoning-first models (o1) on structured STEM, math, and coding.

Key insights

• GPT‑4.5 is not a frontier reasoning model; its strength is deeper understanding of human intent and emotional nuance, validated partly through “vibes testing” with human trainers.
• Hallucination rate dropped to ~37% from GPT‑4o’s 61.2%, making it more trustworthy for general inquiries.
• On the SimpleQA benchmark (single-relation factoid questions), GPT‑4.5 scores 61.9% vs. GPT‑4o’s 38.4%.