
GPT-4o is OpenAI’s omni model that handled text, voice, and images in a single model, and it has now retired from ChatGPT while its API shutdown is set for 14 January 2027.
OpenAI announced GPT-4o on 13 May 2024, and it became the model most ChatGPT users talked to daily. Its text and image features reached free users too. Many older guides still call it the flagship. They are out of date. OpenAI introduced GPT-4o as a step toward more natural human-computer interaction, but newer models have since taken its place.
What GPT-4o is and what “omni” means
The “o” stands for “omni”. Before GPT-4o, talking to ChatGPT chained three separate models: one transcribed speech to text, GPT-3.5 or GPT-4 wrote a reply, and a third turned that reply back into speech. That relay averaged 2.8 seconds with GPT-3.5 and 5.4 seconds with GPT-4, and the middle model never heard tone, laughter, or background noise. GPT-4o replaced the chain with one neural network trained across text, vision, and audio, so one model took in text, audio, images, and video and produced text, audio, and images.
On launch, OpenAI said GPT-4o matched GPT-4 Turbo on English text and code, did better in other languages, and was stronger at vision and audio. In the API it was twice as fast as GPT-4 Turbo, half the price, with five times higher rate limits. A new tokenizer cut token counts sharply, with Gujarati needing 4.4 times fewer tokens. The ChatGPT models guide explains how the lineup around it kept changing after launch.

Where GPT-4o stands now
GPT-4o is gone from ChatGPT. OpenAI announced the retirement in late January and removed GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini from ChatGPT on 13 February. Business, Enterprise, and Edu users kept it in Custom GPTs until 3 April. Old chats continue on newer models. OpenAI said only about 0.1% of users were still picking it each day.
The API tells a different story. The ChatGPT retirement changed nothing in the API at the time, and GPT-4o kept answering API calls after it left the picker. OpenAI’s own model list now marks GPT-4o as Deprecated and the ChatGPT version as Shut down. If you are shopping for a chatbot, the Gemini vs ChatGPT comparison is a better starting point than a retired model.

Voice and vision: the two skills that set it apart
Voice was the skill people noticed first. GPT-4o could answer spoken input in as little as 232 milliseconds, with an average of 320 milliseconds, close to human pace. You could interrupt it, and it heard tone and emotion in your voice. Advanced Voice Mode reached Plus and Team users on 24 September 2024, with preset voices at first.
Vision was the quieter workhorse. You could upload a screenshot, photo, chart, or document page and ask questions about it in the same chat. In the API it understood video without audio, and it worked with a 128,000 token context window. The free AI tools for students roundup lists no-cost options for the same kinds of tasks today.

Image generation inside the chat
On 25 March 2025, OpenAI added native image generation to GPT-4o, replacing DALL-E as the default image maker in ChatGPT. The chat model itself drew the picture, using the conversation and uploaded photos as reference. You could ask for changes in plain language and keep refining the same image, and text inside images came out readable, which suited posters and signs. OpenAI said it could keep 10 to 20 objects straight in a scene. OpenAI’s launch post for 4o image generation describes that conversational editing.
It had clear limits. Tall posters cropped too tightly, small dense text came out unreliable, and non-Latin scripts could go wrong. Detailed images took longer to appear, often up to a minute. The working habits from that era still apply to newer image models: describe the picture precisely, give the exact words you want shown, and ask for one change at a time.

GPT-4o mini and the API shutdown date
GPT-4o mini arrived on 18 July 2024 as the smaller, cheaper member of the family. In ChatGPT it replaced GPT-3.5, and developers used it for high-volume work where speed and cost mattered most. It shares the same retirement schedule.
OpenAI notified developers in October that the snapshots gpt-4o-2024-08-06 and gpt-4o-2024-11-20 leave the API on 14 January 2027. Fine-tuned versions stop the same day as their base models. The original May 2024 snapshot already shut down, and Azure retires the same versions on 14 April 2027. AIVAX transcribed the full deprecation notice, including the replacement model and the fine-tuning cutoff.
The listed replacement is gpt-6-luna, and a straight swap of the model name is a bad plan. Luna is a reasoning model that reasons by default, its tool calling in Chat Completions works only with reasoning effort set to none, and fine-tuning is not supported on it. Run your real prompts against it and fix ones that relied on old behavior before the shutdown date. Developers comparing providers can also read the Claude Code guide.

What replaces GPT-4o and what to do now
If you only ever used GPT-4o inside ChatGPT, there is nothing to do. Your chats moved to the current models in the GPT-6 family, while voice and image features kept working. The model picker no longer lists GPT-4o, and no setting brings it back.
API users have a short checklist. Search your code, plugins, and automations for the model strings gpt-4o and gpt-4o-mini, since the name hides in plugins and no-code tools too. Test gpt-6-luna with real traffic, watch tool calls and output length, and move fine-tuned workloads to a supported path before January. Do the testing early, since the notice period was short.
Alternatives outside OpenAI exist too. The Gemini 3 guide covers Google’s current flagship, the DeepSeek guide covers the low-cost challenger, and the Claude vs ChatGPT comparison puts the two best-known chatbots head to head. Pick by running your own prompts on each.

Frequently asked questions about GPT-4o
What does the “o” in GPT-4o stand for?
The “o” stands for “omni”. GPT-4o was one model trained across text, vision, and audio, so it handled text, speech, images, and video in one model.
Is GPT-4o still available in ChatGPT?
No. OpenAI retired GPT-4o from ChatGPT on 13 February, and it no longer appears in the model picker. Existing chats continue on newer models.
When does GPT-4o leave the API?
The gpt-4o-2024-08-06 and gpt-4o-2024-11-20 snapshots shut down on 14 January 2027, and fine-tuned versions stop the same day. Azure retires the same versions on 14 April 2027.
What replaces GPT-4o?
OpenAI lists gpt-6-luna as the API replacement. In ChatGPT, current models in the GPT-6 family already handle the chats that used to run on GPT-4o.
Was GPT-4o free?
Yes, with limits. Its text and image features reached free users, while Plus users got up to five times higher message limits.
Do GPT-4o image prompts still work on newer models?
The habits carry over: precise descriptions, exact wording for text, reference uploads, and one change at a time.