- New **frontier reasoning + coding model** for “professional work,” released across **ChatGPT**, the **API**, and **Codex**.[5] - Major jump in **computer use / agentic tool use**: on **OSWorld‑Verified** (desktop navigation via screenshots + mouse/keyboard), GPT‑5.4 hits **75.0%**, vs **47.3%** for GPT‑5.2, and even **surpasses human performance (72.4%)**.[5] - Stronger **visual reasoning**: scores **81.2%** on **MMMU‑Pro** without tools (vs 79.5% for GPT‑5.2).[5] - Introduces **tool sea

- This is **OpenAI’s new flagship frontier model** for real‑world professional tasks, replacing GPT‑5.2 Thinking in ChatGPT and the API.[5] - The combination of **state‑of‑the‑art desktop control**, **visual reasoning**, and **tool search** marks a concrete step toward **practical AI agents that can reliably operate software and workflows end‑to‑end**. - Its **token‑efficiency** and integration across ChatGPT, Codex, and enterprise products (e.g., Excel) indicate a new baseline for **product