GPT-4o
By OpenAI
GPT-4o ('omni') is OpenAI's natively multimodal large language model released in 2024, capable of processing and generating text, images, and audio within a single unified model.
Definition
GPT-4o ('omni') is OpenAI's natively multimodal large language model released in 2024, capable of processing and generating text, images, and audio within a single unified model.
Overview
GPT-4o, released in May 2024, was OpenAI's first flagship model designed from the ground up to handle text, vision, and audio natively in one network, rather than stitching together separate specialized models. This unified approach enabled faster, more natural voice conversations in ChatGPT with lower latency than the previous pipeline of separate speech-to-text, language, and text-to-speech models. GPT-4o became the default free-tier model in ChatGPT for a significant period, offering GPT-4-level reasoning and knowledge at meaningfully lower cost and latency than the original GPT-4. It uses the standard Transformer architecture and training pipeline shared across the GPT family: large-scale pretraining, instruction tuning, and RLHF alignment, extended to handle multiple input and output modalities. GPT-4o was later complemented and eventually succeeded as OpenAI's flagship by GPT-4.1 and GPT-5, but it remains widely used via the API for its balance of multimodal capability, speed, and cost, and it introduced many of the real-time voice and vision features that later models built upon.
Key Features
- Natively multimodal — single model for text, vision, and audio
- Low-latency voice conversation mode in ChatGPT
- GPT-4-level reasoning at lower cost and latency than original GPT-4
- Image understanding for photos, screenshots, and documents
- Available via the OpenAI API for developers
- Served as ChatGPT's default free-tier model for an extended period
Use Cases
History
GPT-4o (the "o" stands for "omni") is a multimodal model announced by OpenAI on May 13, 2024 during its Spring Updates event as the new flagship powering ChatGPT. Unlike earlier pipelines that chained separate models, GPT-4o was trained end-to-end as a single network across text, vision, and audio, accepting any combination of those inputs and generating text, audio, and image outputs. It can respond to audio in as little as 232 milliseconds (averaging about 320 milliseconds), close to human conversational latency. At release it matched GPT-4 Turbo on English text and code while being faster and 50% cheaper in the API, and set new state-of-the-art results on multilingual, audio, and vision benchmarks.
Sources
- OpenAI — "Hello GPT-4o" · as of 2026-07-17
- GPT-4o — Wikipedia · as of 2026-07-17