---
title: Hello GPT-4o
author: OpenAI
date: 2024-05-13
original: https://openai.com/index/hello-gpt-4o/
---

We’re announcing GPT‑4o, a flagship model that can reason across audio, vision, and text in real time.

GPT‑4o (“o” for “omni”) accepts any combination of text, audio, image, and video as input and can generate text, audio, and image outputs. It can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds. It matches GPT‑4 Turbo performance on English text and code, improves text performance in non-English languages, and is faster and 50% cheaper in the API. It is also stronger at vision and audio understanding.

# Model capabilities

![Robot on a typewriter](media/robot-writers-block-01.jpg)

![Robot continuing to type](media/robot-writers-block-02.jpg)

![Robot tearing the sheet](media/robot-writers-block-03.jpg)

Prior voice mode used a pipeline of separate models for speech-to-text, text reasoning, and text-to-speech. That process lost information such as tone, multiple speakers, background noise, laughter, singing, and emotion. GPT‑4o is trained end-to-end across text, vision, and audio, so inputs and outputs are processed by one neural network.

# Model evaluations

On traditional benchmarks, GPT‑4o achieves GPT‑4 Turbo-level performance on text, reasoning, and coding intelligence while setting new high watermarks on multilingual, audio, and vision capabilities. The new tokenizer uses fewer tokens across many languages.

# Model safety and limitations

Safety is built into GPT‑4o across modalities through techniques including training-data filtering and post-training refinement. OpenAI evaluated the model for cybersecurity, CBRN, persuasion, and model autonomy, and reports that it did not score above Medium risk in these categories. The model also underwent external red teaming with more than 70 experts in social psychology, bias and fairness, and misinformation.

At launch, text and image inputs and text outputs were made public. Audio outputs were initially limited to preset voices while further safety and technical work continued.