An omni model (sometimes stylised as an "omni" or "any-to-any" model) is an AI system designed to receive and produce multiple modalities - text, audio, images, and video - within a single end-to-end architecture, as opposed to chaining separate specialist models together. The defining characteristic is native cross-modal generation: the model does not just read images and respond in text, but can, in principle, accept speech and produce an image, or accept a video clip and produce a spoken narration, with a unified Latent Space bridging modalities throughout.

Think of it like a universal translator that works not just between languages but between entirely different forms of communication - it hears a melody, watches a gesture, reads a caption, and responds in whichever format best suits the moment.

OpenAI's GPT-4o ("o" for omni) is the most widely cited example; it accepts interleaved text, image, and audio inputs and can produce text and synthesised speech in a single model call rather than routing through separate transcription, reasoning, and text-to-speech services:

# Conceptual GPT-4o audio + image call
response = client.chat.completions.create(
    model="gpt-4o-audio-preview",
    messages=[{
        "role": "user",
        "content": [
            {"type": "input_audio", "input_audio": {"data": audio_b64, "format": "wav"}},
            {"type": "image_url",   "image_url": {"url": image_url}},
            {"type": "text",        "text": "Describe what you hear and see."}
        ]
    }],
    modalities=["text", "audio"]
)

Related concepts include Multimodal (the broader category of models that handle more than one data type), Foundation Model (the large pretrained base that omni models extend), Diffusion (used in some omni architectures for image/video output heads), and Embedding (the shared vector space that ties modalities together during Pretraining).

You will encounter omni models when building real-time voice assistants that can simultaneously process speech and screen content, or when evaluating frontier model APIs that advertise native audio and vision support.