ChatGPT Models and Types

Types of Models

Think of OpenAI’s models less like completely separate tools and more like a family of models with overlapping capabilities, each optimized for different tradeoffs such as speed, cost, and depth of reasoning.

Generative AI (GenAI) models can handle a wide range of tasks, including writing and conversation, analyzing and generating images, transcribing and producing speech, powering recommendations, and performing complex, multi-step analysis. Many modern models are designed to handle multiple types of input and output within a single system, rather than being limited to just one task.

A model that works with only one type of data, such as text, is called unimodal. A model that can work across multiple types of data, like text, images, and audio, is called multimodal. Increasingly, newer models are multimodal, allowing them to understand and generate different forms of content in a unified way, while some unimodal models still exist for highly specialized or efficiency-focused use cases.

Rather than being strictly separated into categories, OpenAI models are often described by their primary strengths or intended use cases:

• Flagship models (e.g., GPT-4o) — General-purpose, multimodal models that balance strong reasoning, speed, and versatility across tasks like chat, image understanding, and audio interaction
• Reasoning-focused models (e.g., o1, o3) — Optimized for deeper, multi-step problem solving, often using more computation to produce more thorough and reliable answers
• Efficient models (e.g., GPT-4o mini) — Smaller, faster, and more cost-effective versions of flagship capabilities, ideal for high-volume or lower-complexity tasks
• Image generation models (e.g., DALL·E) — Designed specifically for creating and editing images from text prompts, though some newer models also integrate vision capabilities directly
• Audio models (e.g., Whisper, text-to-speech) — Focused on speech recognition and audio generation, while some flagship models increasingly support these features natively
• Embedding models (e.g., text-embedding-3) — Convert text into numerical representations for tasks like semantic search, clustering, and recommendation systems

In practice, these categories are not rigid boundaries. Many modern models combine multiple capabilities—such as text, vision, and audio—into a single system, with the main differences coming down to performance, cost, and how much reasoning effort is applied rather than completely separate types of intelligence.

Model Specific Capabilities

Since the creation of ChatGPT there have been many models released. Each model falls into one or more of the various categories or "types" of models mentioned above. Learn more about the different models used by ChatGPT in the table below.

How do you switch between models?

Click on the ChatGPT icon at the top left of the screen. The dropdown will automatically load levels of the latest model. For now, that is GPT 5.3/5.4.

If you'd like to access older models, click "configure." This will provide you with a popup as seen below. Click "latest" to see what other models are available.

Once an individual is well versed with the different models and types of models, they can manually select what model they would like to interact with. Use the chart above to decide which model you should use.

ChatGPT will automatically set the chat to "Instant." This is the latest everyday use model that will be quick and more lightweight. If you aren't satisfied with your answers and would like the chat to dig deeper, switch to "Thinking" or "Pro."