9 ChatGPT

OpenAI

OpenAI’s ChatGPT is arguably the most widely recognised generative AI system available as a commercial product. Launched to the public in November 2022 with its GPT-3 model, ChatGPT allows users to interact with the LLM conversationally. Customers on a paid plan can choose between several models. In March 2023, it progressed to GPT-4, demonstrating remarkable capabilities across various tasks, including coding, analysis, creative writing, and complex problem-solving. GPT-4 introduced multimodal capabilities, allowing it to process text and images. GPT-4o (“o” for “omni”), launched in May 2024, further improved performance and response times, creating a more natural human-computer interaction.

OpenAI’s approach with ChatGPT emphasises broad capability and versatility, resulting in a platform that can handle diverse tasks from routine to complex analysis. The system’s development has prioritised user accessibility and practical utility, making advanced AI capabilities available to users regardless of their technical background.

ChatGPT can be accessed for free or by tiered pricing. The higher your plan type, the more you can use the varying models. Below is a breakdown of the models users can access.

ChatGPT Free

The basic tier offers access to core ChatGPT functionality, albeit with some limitations. Free users receive:

• Access to the GPT-4o mini model
• Limited message capacity during peak usage
• Access to web browsing and image analysis features
• Mobile and web application access

ChatGPT Plus

The individual subscription tier, designed for professional users, provides additional capabilities:

• Priority access to GPT-4o and the latest models
• Increased message limits and faster response times
• Advanced data analysis capabilities
• Image generation through DALL-E integration
• Custom GPT creation and access
• Voice conversation capabilities

ChatGPT Team

Designed for small to medium-sized organisations, Team subscriptions offer:

• Administrative controls and user management
• Higher message limits per user
• Early access to new features and models
• Team workspace for collaboration
• Enhanced data privacy controls
• Usage analytics and reporting
• Dedicated customer support

ChatGPT Enterprise

The premium tier, designed for large organisations and enterprises, includes:

• Unlimited high-speed access to GPT-4o
• Extended context windows up to 128,000 tokens
• Advanced security and compliance features
• Single sign-on (SSO) integration
• Data residency controls
• Dedicated account management
• Advanced analytics and usage insights

Interface

ChatGPT is accessible through OpenAI’s direct chat interface. Interact with this guide below to understand the various functionalities within the ChatGPT interface.

GPT-5

OpenAI’s latest model GPT-5 was released in August 2025 (and is availbake to all users). The release of the model has been described as a “unified system”, which “knows when to respond quickly and when to think longer to provide expert-level responses”.^[1] The ‘unified system’ replaces all earlier models (like GPT‑4.5, GPT‑4o, o3, and the o4-series) and features what OpenAI call a real‑time router. When a user prompts GPT-5, it decides which model to use based on conversation type, complexity, tool needs, and the user’s intent. Watch the following video from OpenAI that introduces and demonstrates GPT-5.

The unified experience also helps users create a prototype application just from a single prompt. Watch the following video which demonstrates how GPT-5 can be used to create a working travel-app prototype in minutes.

GPT-4.5

OpenAI’s released model 4.5 in Feburary 2025). It is suitable for writing and exploring ideas. Watch the following video from OpenAI, which introduces and demonstrates GPT-4.5.

The model builds upon previous models, which were primarily trained using unsupervised learning, and is supplemented with supervised learning and reinforcement learning from human feedback. OpenAI said, ‘Early testing shows that interacting with GPT-4.5 feels more natural. Its broader knowledge base and improved ability to follow user intent make it useful for improving writing, programming, and solving practical problems. We also expect it to hallucinate less.’^[2] Trained with a broad knowledge base, the model is suitable for general-purpose use.

GPT-4.1

In addition, OpenAI released three new models: GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. Currently, these models are only available through their Application Programming Interface (API). Although many users access their technology through the ChatGPT website, many organisations incorporate models directly into their software applications via an API, which acts as a bridge allowing their programs to interact with OpenAI’s systems.

These models feature large context windows (supporting up to 1 million tokens of context) and a refreshed knowledge cutoff of June 2024. OpenAI also reported that these new models outperform their flagship models, GPT-4 and GPT-4 mini, with significant gains in coding and instruction following. Vals reported that these models had a LegaBench accuracy rating of 81.9%, 76.8% and 58.3%, respectively.^[3]

While these models are available via their API (mainly for developers), the improvements observed (e.g., instruction following, coding, and intelligence) will be gradually incorporated into their 4o models.

GPT-4o

The current flagship model of ChatGPT is versatile, accepting both text and image inputs while producing text and image outputs. This model offers the same ‘intelligence’ as GPT-4 but with stronger performance.

Stats on GPT-4o include:

• 128,000 token context window (i.e. an average of a long novel).
• Maximum 16,384 output tokens
• Capable of audio input and output.
• The knowledge cutoff is October 2023
• Achieved 79.8% accuracy on LegalBench benchmark testing

GPT-4o Mini

A lightweight model, 4o mini is OpenAI’s fastest model, perfect for focused tasks. Stats include:

• 128,000 token context window (i.e. an average of a longer novel).
• Maximum 16,384 output tokens
• Capable of text, image and audio input with text, image and audio output.
• The knowledge cutoff is October 2023
• Access is provided on all plans.
• Achieved 76.2% accuracy on LegalBench benchmark testing

GPT-4

OpenAI’s previous flagship model excels in tasks that require creativity and advanced reasoning capabilities. Stats include:

• 128,000 token context window (i.e. an average of a longer novel).
• Maximum 4,000 output tokens
• Text and image input with text and image output.
• Audio input/output.
• The knowledge cutoff is December 2023

The o Series

More recently, OpenAI released the o-series models. These are perfect for tackling complex problems and tasks in research, strategy, coding, math, and science. Created with reinforcement learning, they are designed specifically for tasks that require step-by-step reasoning or logical deduction. They think before answering, producing reasoning steps before responding.

 Watch

The latest o1 model supports both text and image inputs, producing text outputs. The faster o1-mini model currently supports only text inputs and outputs. In 2025, OpenAI released the newer o3-mini model at the same latency and price as the o1-mini. The OpenAI o1 models do not have access to the following advanced tools and features:

• Memory
• Custom instructions
• File Uploads (except images)
• Web browsing
• Discovering and using GPTs
• Voice

GPT-o1

Open AI’s first reasoning model. Stats include:

• 200,000 token context window
• Maximum 100,000 output tokens
• The model supports text and image input, but only produces text outputs
• The knowledge cutoff is October 2023
• Accessible on the ‘plus’ or ‘pro’ plans
• Achieved 77.6% accuracy on LegalBench benchmark testing

GPT-o3 Mini

The most recent OpenAI reasoning model. Stats include:

• 200,000 token context window
• Maximum 100,000 output tokens
• The model only supports text input and output
• The knowledge cutoff is October 2023
• Limited access is provided on the ‘free’ plan, while full access is provided on the ‘plus’ and ‘pro’ plans
• Achieved 70.9% accuracy on LegalBench benchmark testing

OpenAI also has other models, including DALL·E (which can generate and edit images given a natural language prompt) and SORA (which can generate video from a text instruction).