GPT-4o

What it is

GPT-4o is a multimodal model from OpenAI. The "o" in its name stands for "omni" ^[2][6]. OpenAI's system card describes it as an autoregressive omni model: it accepts any combination of text, audio, image and video, and it generates any combination of text, audio and image ^[1]. The system card says it was trained end-to-end, so that a single neural network processes all of these inputs and outputs, rather than passing them between separate systems ^[1].

Background

OpenAI released GPT-4o on 13 May 2024 at a livestreamed event presented by its CTO, Mira Murati ^[2][6]. It succeeded GPT-4 Turbo, and OpenAI's GPT-5 system card names gpt-5-main as its successor ^[4]. Its text and voice capabilities were pre-trained on data up to October 2023 ^[1].

Key features

GPT-4o can respond to audio input in as little as 232 milliseconds, with an average of 320 milliseconds, a pace similar to human response time in conversation ^[3]. On English text and code it matches GPT-4 Turbo, and it improves significantly on non-English text ^[3]. In the API it was faster and 50% cheaper than GPT-4 Turbo; its API price was later listed at $2.50 per million input tokens and $10 per million output tokens ^[2][3]. The context window is 128k tokens ^[1].

At launch the model was free to use in ChatGPT, with higher usage limits for paid subscribers ^[2]. The launch voices were Breeze, Cove, Ember, Juniper and Sky. OpenAI disabled the Sky voice on 20 May 2024 after the actress Scarlett Johansson objected to its resemblance to her voice ^[2].

Its place in the story

GPT-4o was followed by a run of related releases and features. A smaller GPT-4o mini arrived on 18 July 2024, corporate fine-tuning of GPT-4o arrived in August 2024, and the Realtime API was introduced on 1 October 2024 ^[2]. In April 2025 OpenAI rolled back a GPT-4o update after reports that the model had become excessively flattering and agreeable ^[2].

When GPT-5 launched on 7 August 2025, GPT-4o was removed from ChatGPT for non-Pro users. After the backlash, Sam Altman said it would return as an option for Plus users ^[2]. On 29 January 2026 OpenAI announced that GPT-4o would be retired from ChatGPT; it was removed on 13 February 2026, which prompted a "#Keep4o" movement among users ^[2].

In the API, the chatgpt-4o-latest snapshot was shut down on 17 February 2026. The gpt-4o-2024-05-13 snapshot was scheduled for shutdown on 23 October 2026, and the August and November 2024 snapshots on 14 January 2027 ^[5]. GPT-4o therefore shows how a model moves from launch to retirement in ChatGPT and in the API on separate timelines.