GPT-4o

4o · Family: GPT

OpenAI's natively multimodal model (text, image, audio), released May 13, 2024. Real-time voice (~320 ms), 128K context, function calling, 50+ languages.

✓ Active✓ Public accessMultimodalLLMTool-using modelVisionAudioAudio📁 GPT

Context window

128K

tokens

Parameters

Nieujawnione

parameters

Max output

16,384

tokens

Release date

13 May 2024

🏢OpenAIProducer · 2024

Access:APIHostedDeployment:☁ Cloud

Overview

GPT-4o („o” for omni) is OpenAI’s natively multimodal model unveiled on May 13, 2024. A single model handles understanding and generation across text, image, and audio, enabling real-time spoken conversation with an average latency of about 320 ms — close to human response time.

Compared with GPT-4 Turbo it delivers flagship-level quality at a substantially lower price and higher throughput. It supports a 128K-token context window, function calling, Structured Outputs, and more than 50 languages. The model’s knowledge cutoff is October 2023.

Use cases

It excels at voice and text assistants, coding, document and image analysis (OCR, charts), live translation, summarization, and agentic tool-use tasks.

Classification

MultimodalLLMTool-using modelVisionAudioAudio

Family: GPT

Applications

Chatbots Voice assistant Content generation Coding AI coding assistant Q&A / Question answering Summarization Translation

Access & deployment

APIHosted

Cloud

Weights: Closed

Key parameters

📏 Context: 128K

🧩 Parameters: Nieujawnione

✓ Tools · ✓ Fine-tuning

📥 Input: text, image, audio

Platforms

OpenAI API Microsoft Azure AI Foundry OpenRouter

Technical specification

Context window

128K

tokens

Parameters

Nieujawnione

parameters

Max output tokens

16,384

tokens per response

Knowledge cutoff

1 Oct 2023

Knowledge boundary

License

Proprietary (OpenAI)

Hardware requirements

Closed model, available only via the OpenAI API and Azure OpenAI (cloud). No weights available to self-host locally.

Features:✓ Tool use✓ Fine-tuning

Modalities

⬇ Input

textimageaudio

⬆ Output

textimageaudiocode

Capabilities and applications

Native model capabilities

Multimodal understanding

Category: multimodal

Image understanding

Analysing and interpreting the content of images.

Category: vision

Audio understanding

Category: audio

Video Understanding

Category: video

Voice Conversation

Ability to conduct multi-turn real-time voice conversations with context retention and natural speech pacing.

Category: speech

Speech to text

Category: speech

Text to speech

Category: speech

Real-time inference

The model's ability to generate responses with very low latency (>1000 tokens/sec) on specialized inference hardware (e.g. Cerebras WSE), enabling interactive, turn-by-turn collaboration with a human.

Category: coding

Natural conversation

Conducting a conversation with a tone close to human: a warmer voice, empathy in emotional responses, humour, and avoiding the stiff 'AI assistant jargon'. Introduced as a deliberate improvement in GPT-5.1 Instant.

Category: language

Live Translation

Real-time speech translation between multiple languages without interrupting the audio stream.

Category: speech

Coding

Generating, analysing and modifying code in many programming languages. Covers writing functions, debugging, refactoring, code review, and creating tests. Measured by benchmarks such as HumanEval and SWE-bench.

Category: coding

Reasoning

The model's ability to reason logically and solve complex problems.

Category: reasoning

Mathematical reasoning

The model's ability to solve mathematical tasks requiring multi-step reasoning — equations, proofs, combinatorics, geometry, calculus and competition-level problems.

Category: reasoning

OCR

Recognising text within images and documents.

Category: vision

Chart understanding

Reading and interpreting charts, tables and diagrams.

Category: vision

Multilingual

Competence in many natural languages (from a few to over a hundred): understanding, generation, translation, and code-switching within a single conversation. Frontier models support a wide range of languages with comparable quality.

Category: language

Instruction following

Precisely following instructions contained in the prompt: response format, length, style, constraints (e.g. 'reply in six words'). GPT-5.1 significantly improved this capability compared to GPT-5.

Category: language

Long context

Support for large context windows — tens to hundreds of thousands (or millions) of input tokens. Enables analysis of entire codebases, long documents, and many parallel conversations without losing earlier information. GPT-5.1 supports 400,000 tokens.

Category: language

Function Calling

Category: planning

Parallel Tool Calls

Ability to invoke multiple external tools simultaneously while generating a response.

Category: reasoning

Structured output

Producing data in structured formats such as JSON.

Category: structured_generation

Tool use

The model's ability to call external functions, APIs and tools during a conversation: calculator, search engine, code editor, database. The model decides when and how to use a tool and interprets its result.

Category: planning

Streaming output

Category: reasoning

Prompt caching

Cost-performance optimisation: repeated prompt fragments (e.g. system prompt, long documentation) are cached server-side and cheaper in subsequent calls. Significantly reduces cost for applications with long contexts.

Category: other

Interleaved Multimodal Input

Category: reasoning

Application domains

Chatbots Voice assistant Content generation Coding AI coding assistant Q&A / Question answering Summarization Translation Writing assistance Creative writing Research assistance Data analysis Multimodal document understanding Email drafting

Benchmark results

7 benchmarks

MMLU

accuracy · 0-shot CoT

88.7%