Introducing GPT-4o

13 May 202426:13

TLDRIn a groundbreaking presentation, Mira Murati introduces GPT-4o, a new flagship model that brings advanced AI capabilities to everyone, including free users. GPT-4o offers real-time conversational speech, improved text, vision, and audio capabilities, and operates natively across these modalities, reducing latency. The model is designed to be more user-friendly, integrating seamlessly into workflows with a refreshed UI. It also includes features like memory for continuity in conversations and browse for real-time information. GPT-4o's efficiency allows it to be accessible to a wider audience, with over 100 million users already engaging with ChatGPT. The model is also available through the API, offering developers the opportunity to build innovative AI applications. The presentation showcases live demos, including a bedtime story with variable emotional expression, solving a math problem through hints, and real-time translation between English and Italian. The event highlights the potential for GPT-4o to revolutionize human-AI interaction and collaboration.


  • 🌟 The new flagship model, GPT-4o, is introduced, offering GPT-4 intelligence to everyone, including free users.
  • 🚀 GPT-4o is faster and enhances capabilities in text, vision, and audio, marking a significant step in ease of use.
  • 🔍 The model is designed to reduce friction in user interaction, aiming for a more natural and collaborative experience.
  • 🎉 GPT-4o's efficiency allows it to be available to free users, expanding access to advanced AI tools.
  • 📈 GPT-4o can reason across voice, text, and vision natively, improving the real-time experience without latency.
  • 📱 The desktop version of ChatGPT is released, aiming to simplify usage and integration into users' workflows.
  • 🌐 GPT-4o's advanced capabilities are also being made available through the API for developers to build AI applications.
  • 🔧 The team has focused on safety measures to mitigate misuse, especially with real-time audio and vision functionalities.
  • 📉 GPT-4o is 50% cheaper and offers five times higher rate limits compared to GPT-4 Turbo, making it more accessible.
  • 📈 The model supports 50 different languages, reflecting the goal of reaching a wider audience globally.
  • 🔬 Live demos showcased GPT-4o's ability to handle real-time speech, translate languages, and interpret emotions from images.

Q & A

  • What is the main focus of the presentation?

    -The main focus of the presentation is to introduce the new flagship model, GPT-4o, which provides advanced AI capabilities to everyone, including free users, and to showcase its various features and improvements over previous models.

  • What are the key improvements in GPT-4o over its predecessor?

    -GPT-4o offers faster performance, enhanced capabilities across text, vision, and audio, and a more natural and efficient user experience. It also brings these advanced tools to free users and has improved real-time responsiveness and emotion detection in voice interactions.

  • How does GPT-4o handle real-time audio interactions?

    -GPT-4o natively reasons across voice, text, and vision, which allows for real-time conversational speech without the latency issues that were present in previous models. It can also detect emotions and respond accordingly.

  • What new features are available for users in the GPT store?

    -Users can now utilize custom ChatGPT experiences created by others, which are available in the store. Additionally, they can use vision to upload and discuss content containing both text and images, and leverage memory for continuity across conversations.

  • How does GPT-4o support developers through its API?

    -GPT-4o is available through the API, allowing developers to build and deploy AI applications at scale. It offers faster processing, is 50% cheaper, and provides five times higher rate limits compared to GPT-4 Turbo.

  • What are the challenges that GPT-4o presents in terms of safety?

    -GPT-4o presents new safety challenges due to its real-time audio and vision capabilities. The team has been working on building in mitigations against misuse and collaborating with various stakeholders to ensure the technology is introduced safely.

  • How does GPT-4o assist with mathematical problems?

    -GPT-4o can help users solve mathematical problems by providing hints and guiding them through the problem-solving process. It can also recognize and respond to equations written on paper by the user.

  • What is the significance of GPT-4o's ability to understand and generate responses in different languages?

    -GPT-4o's multilingual capabilities allow it to function as a real-time translator, which can be particularly useful for communication between speakers of different languages, thus making the technology more inclusive and accessible globally.

  • How does GPT-4o enhance the user experience with its vision capabilities?

    -GPT-4o can analyze and understand visual content such as screenshots, photos, and documents. It can start conversations based on this content, providing a more interactive and integrated experience for the user.

  • What are the future plans for GPT-4o in terms of deployment and accessibility?

    -The team plans to roll out GPT-4o's capabilities to all users over the next few weeks. They are also working towards the next big technological advancement, with updates to follow in due course.

  • How does GPT-4o's memory feature improve the functionality of ChatGPT?

    -The memory feature allows GPT-4o to maintain continuity across all conversations, making it more useful and helpful by providing a sense of context and history in interactions.



🚀 Introduction to ChatGPT's New Features

Mira Murati opens the presentation by expressing gratitude to the audience and outlining the three main topics of the day. The focus is on making AI tools, specifically ChatGPT, widely accessible and reducing barriers to entry. The release of the desktop version of ChatGPT is announced, promising a more natural and user-friendly experience. The highlight is the launch of GPT-4o, a new model that brings advanced AI capabilities to all users, including those using the free version. Live demos are promised to showcase the model's capabilities, and the mission to provide advanced AI tools for free is emphasized. The presentation also mentions the recent changes to make ChatGPT more integrated into users' workflows and the simplification of the user interface.


🎉 GPT-4o: Expanding Access to Advanced Tools

The second paragraph delves into the excitement surrounding the release of GPT-4o, which is made available to all users, including those who are not paying subscribers. The paragraph discusses the importance of making AI tools accessible for various applications like work, learning, and content creation. It outlines the new features that GPT-4o brings to the table, such as real-time conversational speech, vision capabilities, and improved language support. The paragraph also addresses the challenges of ensuring safety and mitigating misuse as the technology advances, highlighting the collaboration with various stakeholders to navigate these issues.


🤖 Real-Time Interaction and Emotional Response

In the third paragraph, Mark Chen and Barrett Zoph join the stage to demonstrate GPT-4o's capabilities. They showcase the real-time conversational aspect of GPT-4o, which allows for interruptions and immediate responses without lag. The model's ability to detect and respond to emotional cues in speech, such as breathing rate, is highlighted. Additionally, the versatility of GPT-4o's voice generation is demonstrated, as it can adopt different styles, from a storytelling narrative to a robotic voice, and even sing, adding a layer of expressiveness to interactions.


🧮 Solving Linear Equations with ChatGPT

The fourth paragraph features an interactive session where Barrett Zoph solves a linear equation with the help of ChatGPT. The conversational AI provides hints and guidance, demonstrating its ability to assist with mathematical problems. The practical applications of solving linear equations in everyday scenarios are discussed, emphasizing the relevance of mathematics in various aspects of life, from business to cooking. The paragraph also touches on the model's ability to understand and respond to written expressions, as seen when it correctly identifies a handwritten equation.


📈 Analyzing Data and Coding Assistance

The fifth paragraph demonstrates ChatGPT's ability to assist with coding and data analysis. Barrett Zoph shares a piece of code with ChatGPT, which then provides a concise explanation of its functionality. The function in question applies a rolling average to temperature data, smoothing out fluctuations. The paragraph also shows ChatGPT's vision capabilities as it analyzes a plot generated from the code, offering insights into temperature trends and annotating significant weather events.


🌐 Language Translation and Emotional Analysis

The sixth and final paragraph of the script features live audience interactions. ChatGPT is asked to perform real-time translation between English and Italian, showcasing its linguistic capabilities. It also attempts to analyze emotions based on a selfie provided by Barrett Zoph, adding a layer of interactivity and engagement with the audience. The paragraph concludes with a look towards future updates and a thank you note to the team and audience for their support.




GPT-4o is a new flagship model of AI technology introduced in the video. It is described as providing GPT-4 intelligence but in a much faster manner and with improvements across text, vision, and audio capabilities. The term is central to the video's theme as it represents a significant advancement in AI, aiming to make interactions with machines more natural and efficient. In the script, GPT-4o is shown to handle real-time conversations, understand emotions, and even translate languages, showcasing its versatility and advanced capabilities.

💡Real-time conversational speech

Real-time conversational speech refers to the ability of GPT-4o to engage in immediate back-and-forth dialogue without any noticeable lag. This feature is crucial for making interactions with AI feel more natural and human-like. In the script, it is demonstrated when Mark Chen has a real-time conversation with GPT-4o, asking for help with nerves and receiving immediate feedback.

💡Voice mode

Voice mode is a feature that allows users to interact with GPT-4o using spoken language. It is mentioned in the context of improvements over previous models, where GPT-4o allows for interruption and real-time responses. The script illustrates this with a live demo where GPT-4o responds to breathing exercises and storytelling, indicating its advanced voice recognition and processing abilities.

💡Vision capabilities

Vision capabilities pertain to GPT-4o's ability to process and understand visual information, such as images or text within images. This is showcased in the script when Barrett Zoph writes a math problem on paper, and GPT-4o is able to see and assist with solving it. The feature is significant as it expands the ways users can interact with the AI, making it more versatile and useful in various contexts.


Memory, in the context of GPT-4o, refers to the AI's capacity to retain information from previous interactions, allowing it to maintain continuity in conversations. This feature is highlighted as making GPT-4o more useful and helpful by providing a sense of context and history. An example from the script is not provided, but the concept is integral to the narrative of creating a more collaborative and personalized AI experience.


The browse capability allows GPT-4o to search for real-time information during a conversation. This feature is important for providing up-to-date and relevant information to users. In the script, it is mentioned as one of the advanced tools that will be available to all users, enhancing the functionality and utility of GPT-4o.

💡Advanced data analysis

Advanced data analysis is a feature that enables GPT-4o to process and analyze complex data, such as charts and statistical information. This is demonstrated in the script when GPT-4o is described as being able to upload and analyze data, providing users with insights and answers. The feature is significant for its potential applications in various fields that require data interpretation.

💡Language support

Language support refers to GPT-4o's ability to function in multiple languages, with the script mentioning improvements in 50 different languages. This is essential for making the AI accessible and useful to a global audience. The script illustrates this with a live demo where GPT-4o translates between English and Italian in real-time.


API, or Application Programming Interface, is a set of protocols and tools that allows developers to access the functionality of GPT-4o and integrate it into their own applications. The script mentions that GPT-4o will be available through the API, which is significant as it enables developers to create a wide range of AI-driven applications and services.

💡Safety and misuse

Safety and misuse are important considerations when introducing advanced AI technologies like GPT-4o. The script discusses the challenges of ensuring that the technology is used responsibly and does not harm users or society. The team behind GPT-4o is said to be working on building in mitigations against misuse, which is crucial for the ethical deployment of AI.

💡Live demos

Live demos are practical demonstrations of GPT-4o's capabilities shown during the presentation. They serve to illustrate the real-world applications and potential of the technology. In the script, live demos include real-time speech translation, solving math problems, and interacting with visual content, which help to convey the message that GPT-4o is a versatile and advanced AI model.


GPT-4o is a new flagship model that brings GPT-4 intelligence to everyone, including free users.

The desktop version of ChatGPT is being released for broader availability and ease of use.

GPT-4o is faster and improves capabilities across text, vision, and audio.

GPT-4o allows for real-time conversational speech, with no lag and the ability to interrupt the model.

The model can understand and respond to emotions and changes in tone of voice.

GPT-4o can generate voice in a variety of styles, offering a wide dynamic range.

The model can solve linear equations and provide hints to guide users through the process.

GPT-4o can understand and describe code, making it a useful tool for developers.

The model can see and interpret visual content, such as plots and graphs.

GPT-4o can perform real-time translation between English and Italian.

The model can analyze emotions based on a person's facial expression in a photo.

GPT-4o's efficiencies allow advanced tools to be available to all users, not just paid subscribers.

The GPT store allows users to create and share custom ChatGPT experiences.

GPT-4o's memory feature provides continuity across all conversations.

The browse feature allows users to search for real-time information during conversations.

Advanced data analysis tools are integrated, allowing users to upload and analyze charts.

GPT-4o has improved quality and speed in 50 different languages, expanding its global accessibility.

Paid users will continue to have up to five times the capacity limits of free users.

GPT-4o is also available through the API for developers to build AI applications.