GPT-4o

GPT-4o, released on 13 May 2024, was OpenAI's first genuinely end-to-end multimodal model. Earlier voice conversation in ChatGPT was a pipeline of three separate systems — speech recognition, a text model, speech synthesis — which lost tone, background sound and the ability to be interrupted. GPT-4o processed text, audio and images in a single network, answering audio in as little as 232 milliseconds and averaging around 320 ms, close to the pace of human conversation. The "o" stands for omni. Alongside the interaction change, it matched GPT-4 Turbo on English text and code while being roughly twice as fast and half the price, and it improved markedly on non-English languages and on vision and audio understanding. Its context window was 128,000 tokens with a knowledge cutoff in October 2023, and — as with GPT-4 — OpenAI disclosed neither parameter count nor architecture. GPT-4o also became the model people grew attached to. When OpenAI made GPT-5 the default in ChatGPT in August 2025 and removed the older model, the reaction was strong enough that GPT-4o was restored for paying subscribers within days — the first time user sentiment visibly reversed a model retirement. Its actual retirement came in stages: removed from ChatGPT on 13 February 2026, kept in Custom GPTs for business, enterprise and education customers until 3 April 2026, with the chatgpt-4o-latest API endpoint retired on 16 February 2026. A smaller sibling, GPT-4o mini, arrived in July 2024 and for a long stretch was the cheapest capable model on the market, which made it the default choice for high-volume production work rather than for demos.

▸Videos

No videos yet.

▸Reviews