Updated October 15, 2026: This article revisits GPT-4o’s May 2024 launch. The historical account and the dated status update below serve different purposes. This is a review of published documentation; no product or API testing was performed.

About the cover: The retained artwork carries this article’s original 2024 headline. It is an archival illustration, rather than a current ChatGPT interface or evidence of a hands-on test.

What GPT-4o introduced in May 2024

On May 13, 2024, OpenAI introduced GPT-4o, with “o” standing for “omni.” It described a model trained across text, vision and audio.

The important question for someone reading the announcement was which parts they could actually use. A demonstration, a limited alpha and a public release provide different kinds of evidence. Reading them separately gives a clearer picture of the launch.

The initial public release accepted text and images and returned text. The API launched with text and vision support. New audio and video access was planned for a small group of trusted API partners in the following weeks.

OpenAI also announced a new GPT-4o Voice Mode alpha for ChatGPT Plus in the coming weeks. That was an official rollout plan, although it did not guarantee immediate access for every account. These distinctions appear in the announcement’s availability section.

How the original ChatGPT rollout worked

OpenAI’s May 2024 product announcement described a gradual rollout to Free, Plus and Team users, with Enterprise availability to follow. Free access had usage limits; Plus users were promised up to five times the message allowance.

The same announcement said the macOS app was starting with Plus users and would expand in the following weeks. A Windows version was planned for later in 2024. Those statements describe the launch period, not today’s app availability.

For present-day access, use current product documentation and your account’s available controls. A 2024 screenshot cannot establish which model is available today. For earlier context, see our history of ChatGPT’s September 2023 voice and image launch.

GPT-4o status as of October 15, 2026

ChatGPT retirement

According to OpenAI’s ChatGPT retirement notice, GPT-4o retired from standard ChatGPT use on February 13, 2026. Business, Enterprise and Edu customers had an extension within Custom GPTs until April 3, 2026. That extension has ended. The notice separately states that this retirement did not change ChatGPT Voice or ChatGPT Images, which use different models.

API models and their separate deadlines

OpenAI’s GPT-4o API model page lists the gpt-4o alias and dated snapshots as deprecated. It describes text and image input with text output. Audio and video are unsupported for this API model.

The API deprecation schedule distinguishes these exact identifiers:

  • chatgpt-4o-latest: shut down on February 17, 2026.
  • gpt-4o-2024-05-13: scheduled to shut down on October 23, 2026.
  • gpt-4o-2024-08-06 and gpt-4o-2024-11-20: scheduled to shut down on January 14, 2027.

As of this update, the October 23 and January 14 deadlines are still upcoming. Deprecation is advance retirement notice; shutdown ends access. Check the exact model ID and its current lifecycle entry before relying on an older integration.

Example: checking three card labels

This fictional example illustrates a task you could assess with an available image-capable assistant. It is not a GPT-4o test or a report of generated output.

Imagine a photo of three handmade greeting cards laid on a table. Each card has a short handwritten label. A focused prompt could be:

“Read the visible labels on these three cards. Mark any word you cannot read confidently. Then suggest one short, cheerful heading for the display, using only the readable labels as context.”

Separate the answer into observations and suggestions. Compare each reported label with the photo yourself. If a word is unclear, provide a sharper crop or type the word. Judge the proposed heading as a creative option, without treating it as another fact extracted from the image.

This gives you a small, checkable task: three labels to verify and one suggestion to consider. Keep names, addresses and other unnecessary personal information out of the photo. The same evaluation approach remains useful even as the available models change.

What the launch still teaches

GPT-4o’s launch is useful historical context for understanding multimodal assistants. When assessing any launch announcement, ask three concrete questions: What was demonstrated? What was released to the public? What could your account actually access? For ongoing use, add a fourth: Does the exact model or feature still have support?