GPT-4o
OpenAI's first "omni-modal" model — text, audio, image in the same network. Real-time voice conversation with ~320 ms latency.
Launched May 2024. The demo with the voice model led to the "Sky" scandal (Scarlett Johansson claimed the voice copied hers without permission). Native image generation (March 2025) replaced DALL·E 3 — Studio Ghibli-style images went viral and took ChatGPT down for hours. Half the price of GPT-4 Turbo. The Realtime API enables voice assistants with natural interruption handling. Successors: GPT-4.5 (February 2025), GPT-5 (summer 2025).