GPT-4o
Era 5 · Generative AI, reasoning, and agents (2022–2026) › Frontier models
- Lineage
- Neurons and networks
- Kind
- Implementation
Who
OpenAI
What happened
OpenAI presents GPT-4o, a single model trained end to end on text, images, and audio. It answers spoken input with a delay similar to that of a human conversation, and it also reaches free ChatGPT users.
Why it mattered
Text, voice, and images in a single model. Spoken conversation with a machine becomes more fluid.
The full story
Before GPT-4o, ChatGPT's voice mode was a chain of three separate models. A response took 2.8 seconds on average with GPT-3.5 and 5.4 with GPT-4, and information such as tone of voice or background noise was lost along the way.
On May 13, 2024, OpenAI presented GPT-4o, with the "o" standing for "omni." It was a single model trained end to end on text, images, and audio, able to take any combination of text, audio, image, and video as input and to produce text, audio, and images. It responded to speech in as little as 232 milliseconds and in 320 on average, similar to a person's response time in conversation.
The rollout was gradual. That day the text and image features began reaching ChatGPT, free users included, and subscribers got usage limits up to five times higher. For developers it was twice as fast as GPT-4 Turbo and cost half as much.
Voice and video came later: OpenAI announced the new voice mode for subscribers in the following weeks, and audio and video in the API, first for a small group of trusted partners.
Videos
Sources
Original source
- Hello GPT-4o (OpenAI, May 13, 2024) (opens in a new tab). OpenAI, 2024
Scholarship and context
How to cite
“GPT-4o.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/gpt-4o
This entry was last reviewed on November 13, 2026.